Accurate prediction of arsenic and lead contamination in karst groundwater using explainable ensemble machine learning

Abstract

Karst aquifers, characterized by high hydraulic connectivity and limited natural attenuation capacity, accelerate the migration of heavy metals in groundwater, posing a significant risk to regional water supply security. Accurate prediction of arsenic (As) and lead (Pb) contamination in such systems remains challenging because of spatial heterogeneity, sparse monitoring networks, and the structural biases inherent in single-model machine learning approaches. To overcome these limitations, we propose a TOPSIS-based regression ensemble (TRE) framework that integrates seven algorithmically diverse base learners, including tree-based, kernel-based, and connectionist models. An entropy-weighted TOPSIS metalearner assigns optimal aggregation weights through multicriteria performance evaluation, effectively mitigating the structural biases of individual models without the overfitting risk of conventional stacking. The framework achieved excellent predictive performance, with R² values of 0.8937 for As and 0.8877 for Pb. Spatial risk mapping indicated that 12.69% and 10.70% of the study area faces high As and Pb risk, with approximately 9.27 million and 10.26 million residents potentially exposed, respectively. Interpretability analysis revealed distinct driving mechanisms for the two contaminants. As occurrence is influenced by both climatic and anthropogenic processes, with precipitation (36.18%) and population density (33.64%) as primary controls. In contrast, Pb distribution is predominantly governed by anthropogenic activities, with population density contributing 59.90%, followed by precipitation (15.10%) and the karst network development index (14.40%). This study provides a rigorous, data-driven framework for identification and spatially targeted management of groundwater heavy metal contamination in karst regions and offers a methodological reference for similar hydrogeological environments worldwide.

Graphical abstract

Keywords

Groundwater / Karst area / Heavy metals / Ensemble machine learning / Feature importance interpretation / Feature interaction

Highlight

	● Ensemble framework overcomes karst complexity achieving R 2 = 0.89 for As and Pb.
	● Karst conduits bypass natural barriers, accelerating urban Pb contamination.
	● In the world’s largest karst area, more than 10 million residents are at risk of arsenic and lead in groundwater.

References

[1] Belgiu M , Drăguţ L . (2016). Random forest in remote sensing: A review of applications and future directions. ISPRS Journal of Photogrammetry and Remote Sensing, 114: 24–31

[2] Cao W G , Fu Y , Cheng Y P , Zhai W H , Sun X Y , Ren Y , Pan D . (2023). Modeling potential arsenic enrichment and distribution using stacking ensemble learning in the lower Yellow River Plain, China. Journal of Hydrology, 625: 129985

[3] Cardenas-Martinez A , Rodriguez-Galiano V , Luque-Espinar J A , Mendes M P . (2021). Predictive modelling benchmark of nitrate vulnerable zones at a regional scale based on machine learning and remote sensing. Journal of Hydrology, 603: 127092

[4] Chen C , Qu W Y , Gao Z Y , Tian Z G , Zhang S Y , Wang P , Zhao F J . (2026). Soil redox status and dissolved organic matter control the biogeochemical transformations of arsenic in paddy soils. Journal of Hazardous Materials, 501: 140737

[5] Chen J , Fan P P , Zhang F , Tan T J , Fang N F , Wu Z Y , Li Z W , Liu Y J . (2024). Distribution mechanisms of soil surface erosion and underground leakage in karst areas of China: a systematic, quantitative review. CATENA, 246: 108466

[6] Chen Y R , Li H , Zhan Z , Zhang J , Chen Y L , Cao X D , Yeh T C J , Barceló D , Weymer B A , Huang Y . et al. (2025). Artificial intelligence modeling for groundwater environments across spatial scales. Environmental Science & Technology, 59(42): 22351–22372

[7] Chu X W , Ding H H , Zhang X M . (2021). Simulation of solute transport behaviors in saturated karst aquifer system. Scientific Reports, 11(1): 15614

[8] de Carvalho H D P , de Oliveira J F L , de A Fagundes R A . (2025). Dynamic selection of ensemble-based regression models: systematic literature review. Expert Systems with Applications, 290: 128429

[9] Deng J Y , Yang G , Yan X Y , Du J Y , Tang Q , Yu C L , Pu S Y . (2024a). Quality evaluation and health risk assessment of karst groundwater in southwest China. Science of the Total Environment, 946: 174371

[10] Deng J Y , Yu J Y , Wang X T , Yu D , Ma H , Wu Y , Yu C L , Pu S Y . (2024b). Spatial distribution and migration characteristics of heavy metals at an abandoned industrial site in the southwest of China. Journal of Hazardous Materials, 480: 136447

[11] Du J H , Luo S L , Samim S . (2025). Data-driven modeling of syngas yield from biomass pyrolysis using hybrid tree-based machine learning algorithms. Industrial Crops and Products, 236: 121993

[12] El Khoury A , Somogyi A , Chi Fru E , Saleh F , Chraiki I , Fontaine C , Aubineau J , Rollion-Bard C , Harzhauser M , El Albani A . (2025). A battle against arsenic toxicity by Earth’s earliest complex life forms. Nature Communications, 16(1): 4388

[13] Feng Y W , Kang F X , Guo Y L , Li Z F , Liu B T , Bu H Z , Xia C Y , He Q C . (2024). Evolution of hydrochemistry and water quality of karst groundwater under the effects of intense anthropogenic activities. Journal of Environmental Management, 371: 123059

[14] Ford D, Williams P (2007) Karst Hydrogeology and Geomorphology. Hoboken: John Wiley & Sons

[15] Goldscheider N , Chen Z , Auler A S , Bakalowicz M , Broda S , Drew D , Hartmann J , Jiang G H , Moosdorf N , Stevanovic Z . et al. (2020). Global distribution of carbonate rocks and karst water resources. Hydrogeology Journal, 28(5): 1661–1677

[16] Gulati S , Bansal A , Pal A . (2025). Estimating total dissolved solids in groundwater using machine learning models. Natural Resources Research, 34(3): 1623–1644

[17] Guo Z H , Cao J , Xu R , Zhang H L , He L L , Gao H B , Zhu L N , Jia M Y , Yang Z H , Xiong W P . (2024). Novel photoelectron-assisted microbial reduction of arsenate driven by photosensitive dissolved organic matter in mine stream sediments. Environ-mental Science & Technology, 58(50): 22170–22182

[18] Hu H X , Zheng H G , Liu F P , Ding Z Y , Wang Z S , Peng Y H , Zhang D , Zhang Y K , Zheng Y N , Ding A Z . (2024). Heavy metal contamination assessment and source attribution in the Vicinity of an iron slag pile in Hechi, China: integrating multi-medium analysis. Environmental Research, 263: 120206

[19] Huang S J , Cai N G , Pacheco P P , Narandes S , Wang Y , Xu W . (2018). Applications of Support Vector Machine (SVM) learning in cancer genomics. Cancer Genomics & Proteomics, 15(1): 41–51

[20] Huang X , Yao R W , Zhang Y H , Li X , Yu Z Y , Guo H Y . (2025). Data-driven prediction modeling of groundwater quality using integrated machine learning in Pinggu Basin, China. Journal of Hydrology: Regional Studies, 62: 102812

[21] Huang Y P , Wang C , Wang Y H , Lyu G , Lin S J , Liu W J , Niu H B , Hu Q . (2024). Application of machine learning models in groundwater quality assessment and prediction: progress and challenges. Frontiers of Environmental Science & Engineering, 18(3): 29

[22] Inglis A , Parnell A , Hurley C . (2023). vivid: An R package for variable importance and variable interactions displays for machine learning models. The R Journal, 15(2): 344–361

[23] Ji H J , Zhang J , Zhao Y L , Huang H , Ma Y Q , Liang D J , Chen F N , Huo H X , Wang S F , Xie T . (2025). Heavy metal pollution migration and its ecological impact on microbial communities in the karst region of Guangxi. Scientific Reports, 15(1): 14750

[24] Jiang F , Li Q , Sun G T , Wu Q X , Liu S , Jiqin K , Wang R F , Liu H W , Hu W . (2025). Long-term prediction for karst spring discharge and petroleum substances concentration based on the combination of LSTM and Transformer models. Water Research, 274: 123148

[25] Jin X , Wu Q Q , Peñuelas J , Sardan J , Peng Y , Li Z M , Peng X , Heděnec P , Yang Q , Yuan C X . et al. (2025). Climate and anthropogenic activities control the concentrations of copper, zinc, cadmium and chromium in global inland waters. Communications Earth & Environment, 6(1): 520

[26] Jordana S, Batista E (2004). Natural groundwater quality and health. Geologica Acta, 2(2): 175–188

[27] Kumar V , Kedam N , Sharma K V , Mehta D J , Caloiero T . (2023). Advanced machine learning techniques to improve hydrological prediction: a comparative analysis of streamflow prediction models. Water, 15(14): 2572

[28] Li B , Zhang H L , Long J , Fan J , Wu P , Chen M Y , Liu P , Li T . (2022). Migration mechanism of pollutants in karst groundwater system of tailings impoundment and management control effect analysis: gold mine tailing impoundment case. Journal of Cleaner Production, 350: 131434

[29] Li J X , Xie X J , Wang Y X . (2024a). Effects of the groundwater flowing and redox conditions on arsenic mobilization in aquifers of the Datong Basin, northern China. Journal of Hydrology, 636: 131256

[30] Li Y , Wei M L . (2022). Evaluation on adsorption capacity of low organic matter soil for hydrophobic organic pollutant. Journal of Environmental Chemical Engineering, 10(3): 107561

[31] Li Y M , Liu Y , Solangi Y A . (2024b). Analysis of factors and strategies for the implementation of sustainable tourism in a green economic structure in China. Journal of Cleaner Production, 434: 140011

[32] Li Z Q . (2022). Extracting spatial effects from machine learning model using local interpretation method: an example of SHAP and XGBoost. Computers, Environment and Urban Systems, 96: 101845

[33] Liang W Z , Luo S Z , Zhao G Y , Wu H . (2020). Predicting hard rock pillar stability using GBDT, XGBoost, and LightGBM algorithms. Mathematics, 8(5): 765

[34] Liu R T , Wang J G , Zhang Y B , Reimann T , Hartmann A . (2024a). Laboratory and numerical simulations of infiltration process and solute transport in karst vadose zone. Journal of Hydrology, 636: 131242

[35] Liu X , Yue F J , Wong W W , Guo T L , Li S L . (2024b). Unravelling nitrate transformation mechanisms in karst catchments through the coupling of high-frequency sensor data and machine learning. Water Research, 267: 122507

[36] Liu X F , Wang Q F , Zhang Q Z , Yang S J , Liang Y Y , Zhang Y , Li Y , Guan T . (2017). Genesis of the Permian karstic Pingguo bauxite deposit, western Guangxi, China. Mineralium Deposita, 52(7): 1031–1048

[37] Liu Y R , Wang L , Gu K X , Li M . (2022). Artificial Neural Network (ANN)- Bayesian Probability Framework (BPF) based method of dynamic force reconstruction under multi-source uncertainties. Knowledge-Based Systems, 237: 107796

[38] Ming X X , Li X F , Zhang Q Y , Guo H M , Zhang Z , Cao Y S , Shen J X , Zhang C R . (2024). Provenance, chemical weathering, and sedimentary environment of the aquifer sediments: implication for arsenic enrichment in groundwater. CATENA, 239: 107915

[39] Narvaez-Montoya C , Mondragón Bonilla R , Goldscheider N , Mahlknecht J . (2025). Groundwater salinization patterns in the Yucatan Peninsula reveal contamination and vulnerability of the karst aquifer. Communications Earth & Environment, 6(1): 468

[40] Nghiem A A , Stahl M O , Mailloux B J , Mai T T , Trang P T , Viet P H , Harvey C F , van Geen A , Bostick B C . (2019). Quantifying riverine recharge impacts on redox conditions and arsenic release in groundwater aquifers along the Red River, Vietnam. Water Resources Research, 55(8): 6712–6728

[41] O’Connor D , Hou D Y , Ok Y S , Lanphear B P . (2020). The effects of iniquitous lead exposure on health. Nature Sustainability, 3(2): 77–79

[42] Pan J T, Shahbeik H, Shafizadeh A, Rafiee S, Golvirdizadeh M, Nia S A G, Mobli H, Yang Y D, Zhang G L, Tabatabaei M, Aghbashlo M (2024). Machine learning optimization for enhanced biomass-coal co-gasification. Renewable Energy 229: 120772

[43] Peng H , Yang W , Nadine Ferrer A S , Xiong S , Li X Y , Niu G , Lu T T . (2022). Hydrochemical characteristics and health risk assessment of groundwater in karst areas of southwest China: A case study of Bama, Guangxi. Journal of Cleaner Production, 341: 130872

[44] Podgorski J , Berg M . (2020). Global threat of arsenic in groundwater. Science, 368(6493): 845–850

[45] Qu J H , Jiang J Y , Zhang Y F , Li Y P , Yan T G , Guo W , Lu S B . (2022). Soil column experiment and numerical simulation research of Pb migration process in the groundwater fluctuation zone based on HYDRUS. Frontiers in Environmental Science, 10: 989257

[46]

[47] Roberts D R , Bahn V , Ciuti S , Boyce M S , Elith J , Guillera-Arroita G , Hauenstein S , Lahoz-Monfort J J , Schröder B , Thuiller W . et al. (2017). Cross-validation strategies for data with temporal, spatial, hierarchical, or phylogenetic structure. Ecography, 40(8): 913–929

[48] Rodríguez-Lado L , Sun G F , Berg M , Zhang Q , Xue H B , Zheng Q M , Johnson C A . (2013). Groundwater arsenic contamination throughout China. Science, 341(6148): 866–868

[49] Salih A M , Raisi-Estabragh Z , Galazzo I B , Radeva P , Petersen S E , Lekadir K , Menegaz G . (2025). A perspective on explainable artificial intelligence methods: SHAP and LIME. Advanced Intelligent Systems, 7(1): 2400304

[50] Shafizadeh A , Shahbeik H , Rafiee S , Fardi Z , Karimi K , Peng W X , Chen X M , Tabatabaei M , Aghbashlo M . (2024). Machine learning-enabled analysis of product distribution and composition in biomass-coal co-pyrolysis. Fuel, 355: 129464

[51] Shi D X , Zhang B , Liu R , Jiang Z H . (2024). Evaluating close fit in ordinal factor analysis models with multiply imputed data. Educational and Psychological Measurement, 84(1): 171–189

[52] Soltani S S , Cvetkovic V . (2017). Quantifying the distribution of tracer discharge from boreal catchments under transient flow using the kinematic pathway approach. Water Resources Research, 53(7): 5659–5676

[53] Stevanović Z . (2019). Karst waters in potable water supply: a global scale overview. Environmental Earth Sciences, 78(23): 662

[54] Taka M , Sillanpää N , Niemi T , Warsta L , Kokkonen T , Setälä H . (2022). Heavy metals from heavy land use? Spatio-temporal patterns of urban runoff metal loads. Science of the Total Environment, 817: 152855

[55] Wang J X , Shaheen S M , Swertz A C , Liu C S , Anderson C W N , Fendorf S , Wang S L , Feng X B , Rinklebe J . (2024a). First insight into the mobilization and sequestration of arsenic in a karstic soil during redox changes. Environmental Science & Technology, 58(40): 17850–17861

[56] Wang K F , van Genuchten C M . (2025). Commodifying a carcinogen: Critical raw materials from arsenic-laden groundwater. Science Advances, 11(42): eadz5816

[57] Wang S R, Liu X Y, Yuan S H, Bian Y, Wu H, Ye Q (2025a). Artificial intelligence based multispecialty mortality prediction models for septic shock in a multicenter retrospective study. npj Digital Medicine, 8(1): 228

[58] Wang W B , Zhao Y Y , Guan T Z , Qin C X , Chen D N , Jia H F , Zhang X Z , Yu Y H . (2025b). A novel method for hydrology and water quality simulation in karst regions using machine learning model. Ecological Indicators, 178: 113862

[59] Wang X X , Meng L , Hu M Y , Gao L , Lian B . (2024b). The competitive and selective adsorption of heavy metals by struvite in the Pb(II)-Cd(II)-Zn(II) composite system and its environmental significance. Water Research, 250: 121087

[60] Warke M , English M , Padilla C , Gasco L , Leisner W , Datta R , Rao S . (2025). Comprehensive review of in vitro approaches for environmental heavy metal exposure. Frontiers of Environmental Science & Engineering, 19(12): 165

[61] Wu D , Lu J W , Wu J , Li B . (2023). To explore the remediation mechanism and effect of biochar and KH₂PO₄ on soils contaminated with heavy metal ions Hg²⁺, Cd²⁺ and Pb². Journal of Environmental Science and Health Part A-Toxic/Hazardous Substances & Environmental Engineering, 58(13): 1001–1013

[62] Yan Y J , Hu Z Y , Wang L J , Jiang J , Dai Q H , Gan F L , Almaliki A H , Hashim M A , Hussein E E , Ghoneim S S M . (2024). Impact of extreme rainfall events on soil erosion on karst slopes: a study of hydrodynamic mechanisms. Journal of Hydrology, 638: 131532

[63] Yang J , Han Z Y , Yan Y X , Guo G H , Wang L , Shi H D , Liao X Y . (2024). Neglected pathways of heavy metal input into agricultural soil: water-land migration of heavy metals due to flooding events. Water Research, 267: 122469

[64] Yin H R , Zhu Z D , Pan L R , Zhu F Y , Wu X H . (2025). Research on spatial-temporal coupling and driving factors of regional economic resilience and port logistics: empirical evidence from southern Guangxi, China. Systems, 13(7): 524

[65] Zeng G M , Wan J , Huang D L , Hu L , Huang C , Cheng M , Xue W J , Gong X M , Wang R Z , Jiang D N . (2017). Precipitation, adsorption and rhizosphere effect: The mechanisms for Phosphate-induced Pb immobilization in soils-A review. Journal of Hazardous Materials, 339: 354–367

[66] Zeng J W , Liu K , Weng C H , Yan C , Wang X J , Lin X J , Liu N , Qiu J R . (2025). Predicting the fate and source of groundwater PFAS in the pearl river delta region based on machine learning. Environmental Science & Technology, 59(37): 19978–19991

[67] Zhang J W , Sun X H , Wu Q R , Deng J G , Li Z J , Wen M N , Xu L W , Gu Y Z , Han T , Feng L . et al. (2024). Emission characteristics of lead and particulate matter from lead and zinc smelters in China. Journal of Hazardous Materials, 476: 135224

[68] Zhang K , Chai F H , Zheng Z L , Yang Q , Zhong X C , Fomba K W , Zhou G Z . (2018). Size distribution and source of heavy metals in particulate matter on the lead and zinc smelting affected area. Journal of Environmental Sciences, 71(30): 188–196

[69] Zhang S C , Zhan X , Wang J , Xie H H , Quan H , Wang H H . (2025). Lead exposure associated with vitamin D metabolic imbalance mediated by chronic inflammation in children. Frontiers of Environmental Science & Engineering, 19(5): 68

[70] Zhou J , Qiu Y G , Armaghani D J , Zhang W G , Li C Q , Zhu S L , Tarinejad R . (2021). Predicting TBM penetration rate in hard rock condition: a comparative study among six XGB-based metaheuristic techniques. Geoscience Frontiers, 12(3): 101091

[71] Zhou Q , Zhang J J , Zhang S Y , Chen Q , Fan H F , Cao C L , Zhang Y N , Yang Y D , Luo J , Yao Y J . (2025). Groundwater quality evolution across China. Nature Communications, 16(1): 2522

[72] Zhou Y R , Wan X M , Lei M , Chen T B . (2024). Arsenic release during groundwater recharge and effects of coexisting ions in a typical inland basin with high arsenic concentration: modeling and batch experiment. Science of the Total Environment, 950: 175359

[73] Zhu M Z, Cheng R F, Dong P S, Sun H W, Qian K, Shen S, Xie X J (2025). Exploring the groundwater quality index and its pollution sources in southwest karst regions using machine learning models. Journal of Earth Science, doi: 10.1007/s12583-025-0357-4

[74] Zhu Z , Wang J X , Hu M G , Jia L . (2019). Geographical detection of groundwater pollution vulnerability and hazard in karst areas of Guangxi Province, China. Environmental Pollution, 245: 627–633

[75] Zomlot Z , Verbeiren B , Huysmans M , Batelaan O . (2017). Trajectory analysis of land use and land cover maps to improve spatial-temporal patterns, and impact assessment on groundwater recharge. Journal of Hydrology, 554: 558–569

RIGHTS & PERMISSIONS

Higher Education Press 2026