Intermittent river flow forecasting by artificial neural networks

Intermittent stream flow time series consists of zero flows and non-zero flows differing from the rivers having continuous non zero river flow time series. There are stochastic methods in the literature to model the intermittent flow series. In these models different type of probability distributions are employed necessitating the computation of distribution parameters and utilization of different type of goodness of fit tests to represent the stochastic nature of the intermittent flows. In this study artificial neural networks (ANNs) were used to forecast the daily intermittent river flows. Daily mean flow series of a Turkish river was employed for training and testing the ANNs. The forecasted time series were compared with the observed ones. The focus during the evaluation of the forecasted values was given to investigate the potential of ANNs to model the dry periods. It was seen that the dry periods forecasted by ANNs were close to the corresponding observed ones. The ANNs were able to capture the transition between wet and dry periods. Because of the simplicity of application the ANNs could be considered as a reliable alternative to the stochastic models for modelling the intermittent stream flows which carries significance for the water resources projects in and and semi-arid zones.