The employment of polynomial chaos expansion approach for modeling dissolved oxygen concentration in river

Abstract

This article proposes a novel methodology based on polynomial chaos expansions (PCE) for predicting dissolved oxygen (DO) concentration in rivers using four water quality variables as predictors: water temperature, turbidity, pH, and specific conductance. The proposed model is compared to the multilayer perceptron neural network (MLPNN), multilayer perceptron neural network optimized particle swarm optimization (MLPNN_PSO) and the standard multiple linear regression (MLR) with respect to their capabilities for predicting DO. The model results were evaluated using coefficient of correlation (R), Nash–Sutcliffe efficiency (NSE), root mean squared error (RMSE), and mean absolute error (MAE). Using data from more than three stations, operated by the United States Geological Survey (USGS), we demonstrated that that PCE model provides better predicting performance among the different models. Using the four water quality variables led to the best performances of PCE model for modelling DO at all three stations with R and NSE ranging from 0.931 to 0.970, and 0.866 to 0.938, respectively. MLPNN_PSO ranked next with R and NSE which ranged between 0.931 and 0.967, and 0.867 to 0.934, respectively. MLPNN ranked in the third place with R and NSE ranged between 0.921 and 0.966, and 0.849 to 0.931, for the three stations respectively. MLR models led to the worst results with R and NSE ranged between 0.907 and 0.961, and 0.822 to 0.922, respectively. According to the obtained results, PCE model is considered to be a good alternative to the direct measurement of DO concentration in river.