基于机器学习的河南省平原区原生劣质地下水分布预测

doi:10.13745/j.esf.sf.2025.10.33

摘要/Abstract

摘要： 地下水作为全球超20亿人饮用水的重要来源,其水质安全与人类健康和生态系统密切相关。原生劣质地下水(geogenic contaminated groundwater, GCG)是由于地球演化形成的地下水组分超标现象,主要表现为砷、氟、碘含量超标。受地质构造、水文地球化学和人类活动影响,原生劣质地下水分布呈现区域差异和局部突变,其成因机制和防控策略研究对保障水资源安全至关重要。本文以河南省为典型研究区,综合运用Gibbs图等方法解析研究区地下水水化学特征及主控因素,明确原生劣质地下水的分布原因;结合地方病分布数据,探究原生劣质地下水与病区的关联性,并引入机器学习模型实现原生劣质地下水空间分布的精准预测,最终基于上述研究划定地下水健康风险管控区。结果表明:研究区原生劣质地下水集中分布于豫东平原及黄河沿岸区域,且潜水原生劣质组分超标程度显著高于承压水;弱碱性还原环境是原生劣质地下水形成的关键水文地球化学条件,岩石风化溶解与强烈蒸发作用共同主导了特征离子的富集过程;原生劣质地下水分布与地方病区存在一定空间关联性;地下水中砷、氟、碘均表现出明显的空间聚集性,其中高-高(HH)聚类区(即高砷、高氟、高碘地下水叠加区)与机器学习模型识别的高风险区域高度吻合。因此,在濮阳市、新乡市、周口市、开封市及商丘市等重点区域科学划定健康风险保护区,对保障当地居民用水安全具有重要现实意义。

关键词: 砷氟碘, 地下水化学, 健康风险, 地方病, 河南省

Abstract:

As a crucial drinking water source for over two billion people worldwide, groundwater quality is intrinsically linked to human health and ecosystem integrity. Geogenic groundwater contamination (GGC), characterized by excessive levels of arsenic (As), fluoride (F), and iodine (I), originates from natural geological processes. The distribution of GGC, influenced by geological structures, hydrogeochemistry, and anthropogenic activities, exhibits regional patterns with local complexities. Research into its formation mechanisms and control strategies is therefore critical for ensuring water security. Using Henan Province as a case study, this research employed methods including Gibbs diagrams to analyze groundwater hydrochemical characteristics and their controlling factors, thereby identifying the origins of GGC. The correlation between GGC and the distribution of endemic diseases was investigated. Furthermore, machine learning models were introduced to achieve accurate spatial prediction of GGC. Subsequently, health risk control zones were proposed. The results indicate that: (1) GGC in the study area is concentrated in the Eastern Henan Plain and the regions along the Yellow River, with contamination levels in phreatic water being significantly higher than those in confined water; (2) Weakly alkaline and reducing environments represent key hydrogeochemical conditions for GGC formation, where rock weathering and dissolution combined with intense evaporation govern the enrichment of characteristic ions; (3) A spatial correlation exists between GGC distribution and endemic disease areas; (4) Arsenic, fluoride, and iodine in groundwater all exhibit significant spatial aggregation. Notably, the high-high (HH) clusters, indicating areas with co-occurrence of high arsenic, fluoride, and iodine, show strong agreement with the high-risk zones predicted by the machine learning models. Based on these findings, scientifically delineating protection zones in key regions such as Puyang, Xinxiang, Zhoukou, Kaifeng, and Shangqiu cities holds significant practical importance for ensuring local residents’ drinking water safety.

Key words: arsenic, fluorine and iodine, groundwater chemistry, health risk, endemic disease, Henan Province

中图分类号:

P641.2

于福荣, 李蕊, 李志萍, 吴林, 刘中培. 基于机器学习的河南省平原区原生劣质地下水分布预测[J]. 地学前缘, 2026, 33(1): 63-79.

YU Furong, LI Rui, LI Zhiping, WU Lin, LIU Zhongpei. Distribution prediction of natural low-quality groundwater in the plains of Henan Province based on machine learning[J]. Earth Science Frontiers, 2026, 33(1): 63-79.

图/表 16

图1 研究区地理位置及高程图

Fig.1 Geographical location and elevation map of the study area

图2 研究区采样点分布图

Fig.2 Distribution of sampling points in the study area

表1 环境参数

Table 1 Environment parameters

参数类别	具体参数	数据类型	单位/描述
气候气象条件	年平均气温	数值型	℃
	年降水量	数值型	mm/a
	潜在蒸发量	数值型	mm/a
	干燥指数(AI)	数值型	无单位
地表沉积条件	表层岩性类型	类别型	黏土、砂砾等
	土壤渗透系数	数值型	cm/s
	表层有机质含量	数值型	%
	地表盐渍化程度	数值型	无单位
地下水理化特征	pH值	数值型	无单位
	主要阳离子	数值型	mg/L
	主要阴离子	数值型	mg/L
	微量元素	数值型	μg/L
水文地质条件	地下水埋深	数值型	m
水文地质条件	含水层水力传导系数	数值型	m/d
气候气象条件	年平均气温	数值型	℃
	年降水量	数值型	mm/a
	潜在蒸发量	数值型	mm/a

表2 原生劣质组分分类标准(单位:mg/L)

Table 2 Classification standards of Primary Inferior Components (unit: mg/L)

组分	各赋值类别的标准限值/(mg·L^-1)
组分	0	1	2
砷(As)	≤0.01	0.01~0.05	≥0.05
氟(F)	≤1.0	1.0~2.0	≥2.0
碘(I)	≤0.1	0.1~0.50	≥0.50

表3 地下水中砷氟碘浓度表

Table 3 Pollutant concentrations in groundwater

层位	年份	原生劣质组分	最小值	最大值	平均值	标准差	变异系数	原生劣质地下水分布率
潜水	2020	As	0	0.12	0.01	0.01	2.27	14.49%
		F	0.01	3.3	0.72	0.59	0.83	21.98%
		I	0.01	0.8	0.07	0.1	1.36	12.08%
	2021	As	0	0.14	0.01	0.01	2.38	16.91%
		F	0.05	3.4	0.72	0.61	0.85	25.85%
		I	0.05	0.9	0.07	0.09	1.39	7.25%
	2022	As	0	0.16	0.01	0.01	2.81	13.04%
		F	0.05	3.3	0.68	0.58	0.85	22.46%
		I	0.03	1	0.07	0.1	1.4	8.21%
承压水	2020	As	0	0.09	0	0.01	2.55	2.86%
		F	0.05	2.2	0.72	0.52	0.72	32.86%
		I	0.01	1	0.08	0.15	1.77	15.71%
	2021	As	0	0.04	0	0.01	2	4.29%
		F	0.05	2.72	0.75	0.61	0.82	35.71%
		I	0.02	0.9	0.06	0.1	1.61	8.57%
	2022	As	0	0.05	0.01	0.01	2.22	1.43%
		F	0.06	2.56	0.76	0.61	0.8	30.00%
		I	0.05	0.9	0.09	0.15	1.63	10.00%

图3 高砷、高氟和高碘地下水分布情况 a,b,c—分别为2020、2021和2022年潜水;d,e,f—分别为2020、2021和2022年承压水。

Fig.3 Distribution of groundwater with high arsenic, high fluorine and high iodine

图4 研究区地下水Gibbs图 a,b—2020年潜水;g,h—2021年潜水;k,l—2022年潜水;c,d—2020年承压水;i,j—2021年承压水;m,n—2022年承压水。

Fig.4 Gibbs map of groundwater in the study area

图5 研究区地方性饮水型中毒病区与饮用水中砷、氟和碘浓度对比

Fig.5 Comparison of arsenic, fluorine and iodine concentrations in drinking water between endemic drinking water poisoning areas in the study area

图6 SVM(a)、RF(b)、AdaBoost(c)和XGBoost(d)四种模型的混淆矩阵

Fig.6 Confusion matrix of four models

表4 三分类问题的混淆矩阵结构

Table 4 Confusion matrix structure of the three-classification problem

赋值类别	各赋值类别的混淆矩阵
赋值类别	0	1	2
0	TP₀	FN_0-1	FN_0-2
1	FN_1-0	TP₁	FN_1-2
2	FN_2-0	FN_2-1	TP₂

表5 机器模型预测的准确率

Table 5 Accuracy of each model

模型	准确率/%
模型	As	F	I	平均
SVM	81.40	77.48	86.16	81.68
RF	95.87	91.53	96.28	94.56
AdaBoost	89.67	85.74	91.12	88.84
XGBoost	98.14	95.45	97.93	97.17

图7 RF模型分类结果 a—As浓度类别预测结果;b—F浓度类别预测结果;c—I浓度类别预测结果。

Fig.7 Classification results of RF model

图8 XGBoost模型分类结果 a—As浓度类别预测结果;b—F浓度类别预测结果;c—I浓度类别预测结果。

Fig.8 Classification results of XGBoost model

图9 XGBoost模型输出的地下水中砷、氟化物和碘浓度的相关性

Fig.9 Correlation of arsenic, fluoride, and iodine concentrations in groundwater output by the optimal model (XGBoost)

表6 最优模型(XGBoost)输出的原生劣质组分分类全局Moran指数空间自相关性

Table 6 Global Moran index spatial autocorrelation of Primary Inferior Components classification output by XGBoost

离子	Moran’s I	z得分	p值
As	0.108 558	3.777 109	0.000 159
F	0.085 841	2.984 546	0.002 84
I	0.273 991	9.430 531	0

图10 研究区最优预测结果空间关联图的局部指标 a—砷;b—氟;c—碘。HH—高-高聚类;HL—高-低聚类;LH—低-高聚类;LL—低-低聚类。

Fig.10 Local index of spatial correlation graph of optimal prediction results in the study area

参考文献 52

[1]	MALIK A SUGANDH. Groundwater hydro-geochemistry, irrigation and drinking quality, and source apportionment in the intensively cultivated area of Sutlej Sub-basin of main Indus Basin[J]. Environmental Earth Sciences, 2022, 81(18): 456. DOI
[2]	TYOPINE A A, ELOM N I, ALOMAJA F, et al. Assessment of heavy metal contamination and health risks in groundwater of Abakaliki Metropolis, Nigeria[J]. Discover Water, 2024, 4(1): 1-16. DOI
[3]	WANG Y X, LI J X, MA T, et al. Genesis of geogenic contaminated groundwater: As, F and I[J]. Critical Reviews in Environmental Science and Technology, 2021, 51(24): 2895-2933. DOI URL
[4]	纪媛媛. 新疆阿克苏河流域平原区地下水砷氟赋存环境及共富集机理研究[D]. 乌鲁木齐: 新疆农业大学, 2024.
[5]	刘锐平, KHAN T, RAHMAN S, 等. 孟加拉国饮用水砷污染控制策略与方向[J]. 环境工程学报, 2020, 14 (8): 2075-2080.
[6]	HABIB M A, REZA A H M S, HASAN M I, et al. Evaluating arsenic contamination in northwestern Bangladesh: a GIS-Based assessment of groundwater vulnerability and human health impacts[J]. Heliyon, 2024, 10(6): e27917. DOI URL
[7]	KHAN M U, RAI N. Distribution, geochemical behavior, and risk assessment of arsenic in different floodplain aquifers of middle Gangetic Basin, India[J]. Environmental Geochemistry and Health, 2022, 45(5): 2099-2115. DOI PMID
[8]	任宇, 曹文庚, 潘登, 等. 2010—2020年黄河下游河南典型灌区浅层地下水中砷和氟的演化特征及变化机制[J]. 岩矿测试, 2021, 40(6): 846-859.
[9]	ZHU Y, YANG Q, WANG H, et al. A hydrochemical and isotopic approach for source identification and health risk assessment of groundwater arsenic pollution in the central Yinchuan Basin[J]. Environmental Research, 2023, 231(2): 116153. DOI URL
[10]	FENG S, GUO H, SUN X, et al. Relative importance of hydrogeochemical and hydrogeological processes on arsenic enrichment in groundwater of the Yinchuan Basin, China[J]. Applied Geochemistry, 2022, 137: 105180. DOI URL
[11]	NASEEM S, RAFI S. Source of arsenic pollution in southern Indus alluvial plain aquifers, Pakistan[J]. Groundwater for Sustainable Development, 2024, 25: 101157. DOI URL
[12]	ZHANG Y, XIE X, SUN S, et al. Coupled redox cycling of arsenic and sulfur regulates thioarsenate enrichment in groundwater[J]. Science of The Total Environment, 2024, 943: 173776. DOI URL
[13]	LI Y, GUO H, ZHAO B, et al. High biodegradability of microbially-derived dissolved organic matter facilitates arsenic enrichment in groundwater: evidence from molecular compositions and structures[J]. Journal of Hazardous Materials, 2024, 470: 134133. DOI URL
[14]	XIU W, KAI T T, LLOYD J R, et al. Understanding microbial arsenic-mobilization in multiple aquifers: insight from DNA and RNA analyses[J]. Environmental Science & Technology, 2021, 55(22): 15181-15195. DOI URL
[15]	CHEN J, WANG S, ZHANG S, et al. Identifying the hydrochemical features, driving factors, and associated human health risks of high-fluoride groundwater in a typical Yellow River floodplain, North China[J]. Environmental Geochemistry and Health, 2023, 45(11): 8709-8733. DOI PMID
[16]	YUE K, YANG Y, QIAN K, et al. Spatial distribution and hydrogeochemical processes of high iodine groundwater in the Hetao Basin, China[J]. Science of the Total Environment, 2024, 953: 176116. DOI URL
[17]	PATRICK M, SAHU O. Origins, mechanisms, and remedies of fluoride ions from ground and surface water: a review[J]. Chemistry Africa, 2023, 6(6): 2737-2768. DOI
[18]	PODGORSKI J, BERG M. Global analysis and prediction of fluoride in groundwater[J]. Nature Communications, 2022, 13(1): 4232. DOI PMID
[19]	SHAIKH A, KURIAN S, SHENOY D M, et al. Spatial and temporal variation of dissolved iodine in the eastern Arabian Sea[J]. Marine Chemistry, 2023, 257(Suppl C): 104322.
[20]	JOTHIMANI M, ABEBE A, BERHANU G. Application of remote sensing, GIS, and drainage morphometric analysis in groundwater potential assessment for sustainable development in Iyenda River Catchment, Konso Zone, Rift Valley, southern Ethiopia[J]. IOP Conference Series: Earth and Environmental Science, 2022, 982(1): 012032. DOI
[21]	JIANG G, SU C, LIU H, et al. Mobilization and enrichment of geogenic iodine in a floodplain groundwater system: new insights from sources and characterization of dissolved organic matter[J]. Science of the Total Environment, 2024, 956: 177299. DOI URL
[22]	JIANG Z, HUANG M, JIANG Y, et al. Microbial contributions to iodide enrichment in deep groundwater in the North China Plain[J]. Environmental Science & Technology, 2023, 57(6): 2625-2635. DOI URL
[23]	ZHI C, HU B X, CHANG W, et al. Enrichment mechanism of fluoride and iodine in saline groundwater in the lower flood plain of the Yellow River, northern China[J]. Journal of Hydrology, 2023, 621: 129529. DOI URL
[24]	LIU W, QIAN K, XIE X, et al. Co-occurrence of arsenic and iodine in the middle-deep groundwater of the Datong Basin: from the perspective of optical properties and isotopic characteristics[J]. Environmental Pollution, 2023, 329: 121686. DOI URL
[25]	LI J, ZHAO S, JIANG Z, et al. Spatial distribution and formation mechanisms of high-iodine groundwater throughout China[J]. Earth-Science Reviews, 2025, 265: 105134. DOI URL
[26]	王焰新, 曹海龙, 谢先军, 等. 基于树的机器学习方法预测地质成因劣质地下水空间分布[J]. 安全与环境工程, 2022, 29(5): 58-64, 77.
[27]	CAO W, FU Y, CHENG Y, et al. Modeling potential arsenic enrichment and distribution using stacking ensemble learning in the lower Yellow River Plain, China[J]. Journal of Hydrology, 2023, 625: 129985. DOI URL
[28]	CAO W, ZHANG Z, FU Y, et al. Prediction of arsenic and fluoride in groundwater of the North China Plain using enhanced stacking ensemble learning[J]. Water Research, 2024, 259: 121848. DOI URL
[29]	LIU H, LI J, CAO H, et al. Prediction modeling of geogenic iodine contaminated groundwater throughout China[J]. Journal of Environmental Management, 2022, 303(Suppl C): 114249.
[30]	曹文庚, 潘登, 徐郅杰, 等. 河南省滑坡灾害易发性制图研究:多种机器学习模型的对比[J]. 地质科技通报, 2025, 44(1):101-111.
[31]	KNOLL L, MÜLLER S, STUMPP C, et al. Random forest-based prediction of nitrate contamination in groundwater using hydrogeochemical and land-use data[J]. Journal of Hydrology, 2020, 588: 125023.
[32]	ZHAO Y, YANG L, PAN H, et al. Spatio-temporal prediction of groundwater vulnerability based on CNN-LSTM model with self-attention mechanism: a case study in Hetao Plain, northern China[J]. Journal of Environmental Sciences, 2025, 153: 128-142. DOI PMID
[33]	LV S, ZHANG Z, SUN N, et al. Regional groundwater water quality assessment and contamination source identification by a self-organizing map and entropy method in Pinggu Basin, Northeast Beijing[J]. Frontiers in Environmental Science, 2022, 10: 946914. DOI URL
[34]	CAO H, XIE X, SHI J, et al. Evaluating the validity of class balancing algorithms-based machine learning models for geogenic contaminated groundwaters prediction[J]. Journal of Hydrology, 2022, 610: 127933. DOI URL
[35]	王焰新, 李俊霞, 谢先军. 高碘地下水成因与分布规律研究[J]. 地学前缘, 2022, 29(3): 1-10. DOI
[36]	GUO H, LIU C, LU H, et al. Pathways of coupled arsenic and iron cycling in high arsenic groundwater of the Hetao Basin, Inner Mongolia, China: an iron isotope approach[J]. Geochimica et Cosmochimica Acta, 2013, 112(1): 130-145. DOI URL
[37]	CAO H, XIE X, SHI J, et al. Siamese network-based transfer learning model to predict geogenic contaminated groundwaters[J]. Environmental Science & Technology, 2022, 56(15):11071-11079. DOI URL
[38]	AHSAN H, CHEN Y, PARVEZ F, et al. Health effects of arsenic longitudinal study (HEALS): description of a multidisciplinary epidemiologic investigation[J]. Journal of Exposure Science & Environmental Epidemiology, 2006, 16: 191-205.
[39]	LI Y F, WANG D, LIU Y Y, et al. A predictive risk model of groundwater arsenic contamination in China applied to the Huai River Basin, with a focus on the region’s cluster of elevated cancer mortalities[J]. Applied Geochemistry, 2017, 77: 178-183. DOI URL
[40]	MA R, YAN M, HAN P, et al. Deficiency and excess of groundwater iodine and their health associations[J]. Nature Communications, 2022, 13(1): 7354. DOI PMID
[41]	ZHU Y, TONG M, WANG Y, et al. Prevalence of thyroid nodules and its association with water iodine among Chinese men and women[J]. Environmental Research, 2022, 212: 113270. DOI URL
[42]	CAO W, FU Y, REN Y, et al. Arsenic health risk in shallow groundwater of the alluvial plains in the lower Yellow River, China: driving mechanisms of climate change and human activities[J]. Environment International, 2025, 202: 109711. DOI URL
[43]	贾秀阁, 吴冰华, 王唯锦, 等. 河南省浅层原生劣质地下水分布特征研究[J]. 河南科学, 2022, 40(1): 93-97.
[44]	LI X, GUO H, JIAO J J, et al. Modified Gibbs diagram for high-fluoride groundwater classification: a case study in the Loess Plateau[J]. Applied Geochemistry, 2020, 123: 104788. DOI URL
[45]	ADIMALLA N, QIAN H. Hydrogeochemical characterization and groundwater quality assessment in urbanizing regions using Gibbs ratios and pollution indices[J]. Environmental Science and Pollution Research, 2021, 28(45): 63815-63832. DOI
[46]	ZHANG Y, LI P, QIAN H, et al. Hydrochemical evolution and controlling mechanisms of groundwater in the arid Northwest China: insights from Gibbs diagrams and ionic ratios[J]. Journal of Hydrology, 2021, 603: 127045. DOI URL
[47]	KUMAR M, RAMANATHAN A, SINGH V B, et al. Deciphering groundwater salinization in coastal aquifers: an integrated approach using Gibbs plots, ionic deltas, and isotopic tracers[J]. Science of the Total Environment, 2022, 806: 150520. DOI URL
[48]	HE S, WU J. Hydrogeochemical signatures of mining-impacted aquifers: combining Gibbs diagrams with Sr²⁺/Ca²⁺ and δ³⁴S-${\mathrm{SO}}_{4}^{2-}$ ratios[J]. Chemosphere, 2023, 311: 137013.
[49]	GAO Z, LIU J, WANG Z, et al. Ionic ratio evolution in arsenic-fluoride co-contaminated aquifers: implications for redox zonation and remediation[J]. Water Research, 2022, 219: 118543. DOI URL
[50]	PODGORSKI J E, LABHASETWAR P, SAHA D, et al. Global-scale groundwater ion ratio analysis: a machine learning-enhanced Gibbs framework[J]. Nature Water, 2023, 1(5): 345-356.
[51]	原春生, 马嘉俊, 李晓红, 等. 2019年河南省饮水型地方性砷中毒监测结果分析[J]. 中华地方病学杂志, 2023, 42(4): 325-327.
[52]	原春生, 马嘉俊, 闫英洁, 等. 2021年河南省饮水型地方性氟中毒监测结果分析[J]. 中华地方病学杂志, 2023, 42(11): 883-887.