Daily Streamflow Prediction for Khazir River Basin Using ARIMA and ANN Models

Abstract

The present study used both Autoregressive Integrated Moving Average (ARIMA) and Artificial Neural Network (ANN) models for Khazir river basin to simulate the daily flow at Asmawa and Khanis gauge stations. Asmawa station lies on Khazir River while Khanis lies on Gomel River as a tributary of Khazir River. In the stochastic ARIMA model, the Autocorrelation function (ACF) and partial autocorrelation function (PACF) were used to determine how robust the ARIMA model is in predicting the streamflow. In this study, the Akaike Information Criterion (AIC) formula and Bayesian information criterion (BIC) were used to evaluate which model is more accurate. The results of this study showed that models of order ARIMA are (2,0,0)(2,1,0) and (2,0,1)(2,1,0) were found much better than the other models for generating and forecasting daily flow time series for aforementioned stations. Coefficients of determination (R²) were found 0.77 and 0.85 for both Asmawa and Khanis stations, respectively. However, two types of ANN models were used for analyzing the daily flow records of the same two aforementioned stations, Multilayer Perceptron (MLP) and Radial Basis Function (RBF). ANN-MLP model was found to be more accurate than the ANN-RBF for generating and forecasting the daily flow time series as the coefficient of determination provided by ANN-MLP for both stations were 0.83 and 0.85, respectively. In addition, the coefficients of determination produced by the ANN-RBF for both stations were 0.66 and 0.55, respectively. Based on the values of (R²) and (RMSE) obtained in the current work, one can conclude that the ANN-MLP model is the most accurate model among the others in terms of predicting the streamflow for Asmawa station, whereas the performance of both ARIMA and ANN-MLP models for the Khanis station is the same.