Abstract Full Text:PDF Recommended articles Mail to Author How to Cite

¹ Bursa Uludag University, Faculty of Engineering, Environmental Engineering Department, Bursa/Türkiye
² Division of Sustainable Development, College of Science and Engineering, Hamad Bin Khalifa University, Qatar Foundation, Doha/Qatar DOI : 10.4194/TRJFAS30206 Viewed : 162 - Downloaded : 113 Artificial neural networks are important management tools that can be used to manage water pollution. In this study, artificial neural networks (ANNs) were used to predict the water quality of Doğancı dam. The water quality data of the Doğancı dam, from 1990 to 2020, comprised water temperature, alkalinity, turbidity, pH, and concentrations of suspended solids, dissolved oxygen, arsenic, manganese, and iron. Feed-forward neural networks with one input, one hidden, and an output layer were chosen, and the number of nodes was examined. The results indicated that the model incorporating water temperature, alkalinity, pH, and concentration of dissolved oxygen as inputs and trace elements (arsenic, manganese, and iron) as outputs, performed better, having a sufficiently higher correlation-R (between 0.994 and 0.999 for the entire dataset), lower root mean square error-RMSE (ranging from 0.01 to 0.03), and higher Nash-Sutcliffe efficiency-NSE (up to 0.999). Levenberg-Marquardt and resilient backpropagation training algorithms were also compared, with Levenberg-Marquardt's outperforming in the correlation and error metrics for the models tested. The findings provide a useful framework for predicting trace metals in drinking-water reservoirs using long-term monitoring data. This approach supports water-quality assessment, early-warning monitoring, and public-health-oriented management. Keywords : Artificial neural networks, Drinking water dam, Training algorithms, Water quality modeling