Modeling daily dissolved oxygen concentration using modified response surface method and artificial neural network: a comparative study

In the present study, two nonlinear mathematical modeling approaches, namely modified response surface method (MRSM) and multilayer perceptron neural network (MLPNN) were developed and compared for modeling daily dissolved oxygen (DO) concentration. The DO concentration and water quality variables data for several years, available from four stations operated by the United States Geological Survey, were used for developing the two models. The water quality data selected consisted of daily measured river discharge, water pH, specific conductance, water turbidity, and DO. The response surface methodology is modified based on the two steps for calibrating process. In the first regression step, the normalized input data were calibrated based on a linear function and then transferred by an inverse power function. In the second regression step, the input data from first step were used to calibrate a highly nonlinear third-order polynomial function. The accuracy of the proposed nonlinear MRSM is compared with the standard MLPNN using several error statistics such as root-mean-square error, mean absolute error, mean bias error, the coefficient of correlation, the Nash–Sutcliffe efficiency, and the Willmott index of agreement. The results obtained indicate that MRSM model performed best in comparison with the MLPNN for the all four stations.