Hydrological modelling with artificual neural networks

It is very important to make reliable flow forecasts when planning arrangements andapplications in rivers. Especially when the amount of investment required for areservoir promotion is taken into consideration, the reliability in forecasting is ofgreater importance. Traditional flow forecasting methods can be insufficient becauseof the uncertainty and non linear characteristics of the system. Therefore alternativeforecast methods are needed to achieve better predictions.In this study, the forecasting of flows in a river due to the rainfall in its basin and theformer flow observations is investigated. Artificial Neural Networks have been used insimilar forecasting applications in recent years. Having been given the theoreticalbackground on the subject, the Akarçay river basin, a closed catchment in MiddleAnatolian Region, is selected for the case study. Four types of models are planneddue to the parameters such as the placement of rainfall observation stations andobservation intervals. The models are formed by training and testing procedures assuitable to the ANN Methodology. The results are compared with those of multivariable regression analysis. These comparisons are presented in tables and graphs.As a result, it is found that Artificial Neural Networks could be successfully applied inthe flow forecasts by rainfall observations and are capable of giving reliableforecasts.KEY WORDS: Rainfall-runoff, artificual neural networks, multiple regression,hydrological modelling