ANN hybrid model versus ARIMA and ARIMAX models of runoff coefficient

Abstract

• We modeled monthly direct runoff coefficients to forecast a holdout dataset. • Initially, we used univariate ARIMA, multivariate ARIMAX, and ANN models. • We found the applied traditional model performances insufficient. • We developed a new Hybrid approach by using time series decomposition and ANN. • We found that the new generated model is superior, suitable for complicated data. In this study, monthly runoff coefficients of seven southern large basins are calculated and modeled to forecast a holdout dataset by using univariate autoregressive integrated moving average (ARIMA), multivariate ARIMA (ARIMAX), and Artificial neural network (ANN) models. The applied traditional model performances are found insufficient, since the characteristic behaviors of the time series of direct runoff coefficients are very complicated. Therefore, a new Hybrid approach is adopted by using time series decomposition procedure and ANN. ARIMA, ARIMAX, ANN, and Hybrid models are compared with each other. The results indicate that the new generated Hybrid approach can be generalized to boost the prediction capability of ANNs in complicated time series data. It is seen that the new model captures the physical behavior of the direct runoff coefficient time series. The semi-random spikes of the direct runoff coefficient series are approximated sufficiently.