An integrated approach for water quality assessment and pollution source identification using optimized machine learning and water quality index model in a Tidal River of Bangladesh

Faruq, Omur, Hossain, Nahrin Jannat, Sajib, Abdul Majed, Diganta, Mir Talas Mahammad, Moniruzzaman, Md, Olbert, Agnieszka I., & Uddin, Md Galal. (2026). An integrated approach for water quality assessment and pollution source identification using optimized machine learning and water quality index model in a Tidal River of Bangladesh. Journal of Hydrology: Regional Studies, 64, 103215. https://doi.org/10.1016/j.ejrh.2026.103215

Abstract

Study region The Bhairab River is located in the south of Bangladesh. It is an active tidal river that supports a wide range of aquatic environments. Study focus The present research utilized a holistic approach by incorporating the optimized root mean squared water quality index (RMS-WQI) model and machine learning/artificial intelligence (ML/AI) techniques to assess the water quality (WQ) of the Bhairab River. The study utilized four years (2021–2024) of WQ data, including temperature, pH, electrical conductivity, chloride, total solids, dissolved oxygen, and biochemical oxygen demand, from 8 monitoring sites of the Bhairab River. New hydrological insights of the region The results of the RMS-WQI model showed a decreasing trend of water quality index (WQI) scores between 2021 and 2024, with most of the monitoring sites rated as ‘fair’ to ‘poor’ WQ categories, indicating that the majority of WQ indicators failed to meet World Health Organization (WHO 2022) and Environmental Conservation Rules (ECR, 2023) standards. The declining trend of WQI scores was statistically validated by the Friedman test statistic of 21.75 (p-value < 0.05) and the Mann-Kendall Tau value of -1.0 (p-value < 0.05). Moreover, the study utilized eight ML/AI algorithms with the Optuna optimizer, where the Artificial Neural Network (ANN) model demonstrated excellent performance with high accuracy and reliability in predicting WQI scores. In terms of reliability assessment, the ANN-Optuna model showed high effectiveness (Average model efficiency factor: MEF = 0.47; average percentage of relative error index: PREI = 0.39), excellent sensitivity (Average coefficient of determination: R2 = 0.95), and low uncertainty throughout the study period. In addition to assessing WQ trends, the study identified major pollution hotspots along the Bhairab River, where various types of industrial activities, brick kilns, and urban dumping stations were identified as the major sources of pollution around the monitoring sites. In summary, the declining trend of WQ indicated that the Bhairab River was under notable pressure from various point and non-point pollution sources during the study period, which requires site-specific WQ management strategies to protect the Bhairab River ecosystem and living organisms.

Publisher

Elsevier

Rights

CC BY

cb

Collections

Publications