LAUSR

Abstract The need for accurate predictions of water quality in rivers has encouraged researchers to develop new methods and to improve the predictive ability of conventional models. In recent years, artificial intelligence (AI)-based methods have been recognized significantly powerful for this purpose. In this study, the performance of the various types of single and hybrid AI models including artificial neural networks (ANNs), genetic programming (GP), fuzzy logic (FL), support vector machine (SVM), hybrid neuro-fuzzy (NF), hybrid ANN-ARIMA, hybrid genetic algorithm-neural networks (GA-NN), and wavelet-based hybrid models such as wavelet-neural networks (WANN), wavelet-neuro fuzzy (WNF), wavelet-support vector regression (WSVR), and wavelet-linear genetic programming (WLGP) models were investigated for the prediction of water quality in rivers. In this review paper, for each of the models, firstly, a brief introduction is provided. Then some recently published papers are presented to review the performance of the model for modeling water quality in rivers. For this purpose, 51 journal papers that were published from 2000 to 2016 and dealing with the use of the single and hybrid AI models for river water quality prediction were selected. The review of these papers is undertaken in terms of the predictor selection, data normalization, train, and test data division, modeling approaches, prediction time steps, and modeling performance evaluation procedures. The effect of using integrated models to improve the prediction accuracy of the single models was investigated as well. Out of the 51 selected papers, 31 papers (~60% of the entire papers) were published in the past five years. The selected papers have been cited up to 1716 times before 20th February 2016. Among the various modeling techniques, the ANN and WANN models (17 and 7 papers, respectively) were the most widely used single and hybrid models. In the reviewed papers, more attention is given to the modeling of dissolved oxygen (DO) and suspended sediment in rivers. In 23 papers, data with daily time intervals were used for water quality modeling. The present paper covers 13 different single and hybrid AI models. It presents a comprehensive investigation into the application of AI methods for modeling river water quality and offers a critical insight into the use and reliability of the various modeling approaches for modeling diverse water quality measurements.