Abstract Surface water quality in hyper-arid regions is crucial for sustaining agricultural and domestic activities. Increasing human activities, including religious practices and inadequate waste management, pose significant challenges, leading to bacterial contamination and deteriorating water quality. This situation demands urgent attention. This study investigates the hydro-chemical properties and contamination levels of surface water, focusing to examine the water quality of Baitarani River Basin, Odisha, through frequent field studies, conducted along the thirteen surveyed stations and the use of multivariate statistical methods, such as the Pearson correlation and Principal Component Analysis (PCA). Over a 9-year (2014–2023) period, a total of 14 samples were analysed for physicochemical parameters, including total dissolved solids (TDS), electrical conductivity (EC), pH, turbidity, coliform, alkalinity, and major ions (Ca²⁺, Mg²⁺, Na⁺, K−, Cl⁻, SO₄²⁻, HCO₃⁻, NO3−). Water quality was evaluated across thirteen locations using the Surface (Su) Water-Weighted Water Quality Index (WQI) to determine its suitability for human consumption and restricted irrigation. The calculated Su-WQI values ranged from 43 to 223, signifying excellent to extremely poor water quality. The findings suggest that, while five (38.46%) locations along the river are suitable for drinking and irrigation, and in contrast, areas near wastewater discharge i.e., 8 locations classified 61.54% of the samples renders very poor/extremely poor, highlighting significant salinity, coliform, turbidity, TDS, bicarbonate, and sulphate pollution, thus emphasizing the urgent need for sustainable surface water management and effective remediation strategies. Geochemical diagrams (Gibbs and Piper) suggests ions, that are mostly discharged in water as a result of rock-water contact and carbonate weathering. Pearson’s coefficient is used to classify the relationships between parameter pairs according to the strength of correlation and quantify the degree of linkage. In addition, multivariate statistical tools such as Principal Component Analysis (PCA) revealed five PCs, demonstrating 86.65% of the total variance. Utilizing Machine Learning (ML) techniques, the surface water quality parameter was evaluated for training and testing the models using a classifier attribute evaluator. Subsequently, Model performance was examined using metrics including accuracy, precision, recall, and false positive rate (FPR) that were derived from the confusion matrix. With a low FPR and mean absolute error (MAE) of 75.17% and 0.225, a high accuracy of 82.44%, a precision of 84.29%, and a recall of 96.59%, Optimized Forest performs better than any other classifier. The Optimized Forest model can be utilized in surface water monitoring locations to test fresh water samples for the classification of water quality indicators. As a result, by emphasizing the reuse of water for urban greening and sustainable development, the study aligns with Odisha’s Vision commitment to resource conservation, ecological balance, and creating a greener, more sustainable urban environment. The findings offer critical insights for policymakers and stakeholders in developing comprehensive water resource management frameworks. Graphical abstract Based on the graphical snapshot, this study was conducted to determine the variation in Hydro-chemical facies, understand the evolution of Hydro-chemical processes and to assess the suitability of surface water for both agricultural and drinking purposes. This work captures complex relationships between socioeconomic and environmental variables and surface water assessment in Baitarani Basin, Odisha. Surface water sampling was carried out considering both spatial and chemical controls to understand the underlying mechanisms operating within a dynamic alluvial hydrogeochemical environment. Hydro-chemical parameters including pH, EC, major cations, and anions were analysed following standardized procedures outlined by the American Public Health Association. Detected water samples were collected from relevant locations and analysed for various physicochemical parameters. Multiple indices, including Pearson Correlation, Principal Component Analysis, and Machine Learning (ML) approaches, were utilized in the models. Pearson’s coefficient to quantify the degree of association and categorizes the relationships between parameter pairs based on the magnitude of correlation. Principal Component Analysis (PCA) revealed five PCs, demonstrating 86.65% of the total variance. Piper’s and Gibb’s plots revealed that surface water belong to Ca-Cl2 type, followed by mixed Ca-MgCl2, and Ca-(HCO3)2 type water, illustrating the presence of permanent hardness and the dominance of alkaline earths over alkali, with a nondominant cation type. This plot also demonstrated the dominance of reverse ion exchange water having permanent hardness in majority of the samples over recharging water with temporary hardness. As per ML observation, parameters obtained from the confusion matrix, such as accuracy, precision, recall, and FPR, were used to analyses the performance of models. Among all models, Optimized Forest outperforms other classifier as it has a high accuracy of 82.44%, a precision of 84.29%, recall of 96.59%, and a low FPR and MAE of 75.17% and 0.225. The findings offer critical insights for policymakers and stakeholders in developing comprehensive water resource management frameworks.

Prediction of Urban Surface Water Quality Scenarios Using Water Quality Index (WQI), Multivariate Techniques, and Machine Learning (ML) Models in Water Resources, in Baitarani River Basin, Odisha: Potential Benefits and Associated Challenges

期刊论文
发表日期：

2025年4月9日

查看全文

Abhijeet Das

摘要

Abstract

Surface water quality in hyper-arid regions is crucial for sustaining agricultural and domestic activities. Increasing human activities, including religious practices and inadequate waste management, pose significant challenges, leading to bacterial contamination and deteriorating water quality. This situation demands urgent attention. This study investigates the hydro-chemical properties and contamination levels of surface water, focusing to examine the water quality of Baitarani River Basin, Odisha, through frequent field studies, conducted along the thirteen surveyed stations and the use of multivariate statistical methods, such as the Pearson correlation and Principal Component Analysis (PCA). Over a 9-year (2014–2023) period, a total of 14 samples were analysed for physicochemical parameters, including total dissolved solids (TDS), electrical conductivity (EC), pH, turbidity, coliform, alkalinity, and major ions (Ca²⁺, Mg²⁺, Na⁺, K⁻, Cl⁻, SO₄²⁻, HCO₃⁻, NO₃⁻). Water quality was evaluated across thirteen locations using the Surface (Su) Water-Weighted Water Quality Index (WQI) to determine its suitability for human consumption and restricted irrigation. The calculated Su-WQI values ranged from 43 to 223, signifying excellent to extremely poor water quality. The findings suggest that, while five (38.46%) locations along the river are suitable for drinking and irrigation, and in contrast, areas near wastewater discharge i.e., 8 locations classified 61.54% of the samples renders very poor/extremely poor, highlighting significant salinity, coliform, turbidity, TDS, bicarbonate, and sulphate pollution, thus emphasizing the urgent need for sustainable surface water management and effective remediation strategies. Geochemical diagrams (Gibbs and Piper) suggests ions, that are mostly discharged in water as a result of rock-water contact and carbonate weathering. Pearson’s coefficient is used to classify the relationships between parameter pairs according to the strength of correlation and quantify the degree of linkage. In addition, multivariate statistical tools such as Principal Component Analysis (PCA) revealed five PCs, demonstrating 86.65% of the total variance. Utilizing Machine Learning (ML) techniques, the surface water quality parameter was evaluated for training and testing the models using a classifier attribute evaluator. Subsequently, Model performance was examined using metrics including accuracy, precision, recall, and false positive rate (FPR) that were derived from the confusion matrix. With a low FPR and mean absolute error (MAE) of 75.17% and 0.225, a high accuracy of 82.44%, a precision of 84.29%, and a recall of 96.59%, Optimized Forest performs better than any other classifier. The Optimized Forest model can be utilized in surface water monitoring locations to test fresh water samples for the classification of water quality indicators. As a result, by emphasizing the reuse of water for urban greening and sustainable development, the study aligns with Odisha’s Vision commitment to resource conservation, ecological balance, and creating a greener, more sustainable urban environment. The findings offer critical insights for policymakers and stakeholders in developing comprehensive water resource management frameworks.

Graphical abstract

Based on the graphical snapshot, this study was conducted to determine the variation in Hydro-chemical facies, understand the evolution of Hydro-chemical processes and to assess the suitability of surface water for both agricultural and drinking purposes. This work captures complex relationships between socioeconomic and environmental variables and surface water assessment in Baitarani Basin, Odisha. Surface water sampling was carried out considering both spatial and chemical controls to understand the underlying mechanisms operating within a dynamic alluvial hydrogeochemical environment. Hydro-chemical parameters including pH, EC, major cations, and anions were analysed following standardized procedures outlined by the American Public Health Association. Detected water samples were collected from relevant locations and analysed for various physicochemical parameters. Multiple indices, including Pearson Correlation, Principal Component Analysis, and Machine Learning (ML) approaches, were utilized in the models. Pearson’s coefficient to quantify the degree of association and categorizes the relationships between parameter pairs based on the magnitude of correlation. Principal Component Analysis (PCA) revealed five PCs, demonstrating 86.65% of the total variance. Piper’s and Gibb’s plots revealed that surface water belong to Ca-Cl₂ type, followed by mixed Ca-MgCl₂, and Ca-(HCO₃)₂ type water, illustrating the presence of permanent hardness and the dominance of alkaline earths over alkali, with a nondominant cation type. This plot also demonstrated the dominance of reverse ion exchange water having permanent hardness in majority of the samples over recharging water with temporary hardness. As per ML observation, parameters obtained from the confusion matrix, such as accuracy, precision, recall, and FPR, were used to analyses the performance of models. Among all models, Optimized Forest outperforms other classifier as it has a high accuracy of 82.44%, a precision of 84.29%, recall of 96.59%, and a low FPR and MAE of 75.17% and 0.225. The findings offer critical insights for policymakers and stakeholders in developing comprehensive water resource management frameworks.