Use of artificial intelligence techniques in estimating river flows

As the demand for water increases, efficient planning and management of water resources have required more reliable estimates for the yields of water resources. River flow estimates based on past river flows and the rainfalls recorded in the basin have been undertaken conventionally by various stochastic time series approaches such as autoregressive (AR), moving average (MA) and autoregressive moving average (ARMA). Artificial intelligence (AI) techniques have been used for some time to replace such traditional techniques due especially to their capability of modeling complex nonlinear processes. In this study, various AI techniques are used to predict river flows from the past flows and from the values of upstream stations. The AI techniques used include feed forward back propagation neural networks, generalized regression neural network, fuzzy logic. The classical multiple regression analysis is also used for comparison. A series of models are constructed to predict river flows in the case of Birs River in Switzerland. The comparisons of the performance of the models used are accomplished based on selected performance criteria such as mean square error, determination coefficient and efficiency coefficient. The study has shown that AI techniques are quite capable of modeling river flows and providing reliable estimates for them.