Artificial neural network applied to the forecast of streamflow in the Pianco River Basin

Streamflow forecasting in a water system is one of the techniques used to reduce the impact of the uncertainties of the climate on administration of the water resources. That technique can be considered as one of the principal challenges related to the integrated knowledge of the climatology and of the hydrology of the river basin. The aim of this work was to model the non-linear relationship between rainfall and streamflow in the Pianco River Basin, in the Paraiba semiarid, using the technique of Artificial Neural Networks (ANN). Here the ability of ANN was evaluated to model the rainfall-runoff process on a monthly basis. During training of the ANN, the network architecture and weights initialization influence were considered. At the end of the training the best architecture was chosen, to model the streamflow monthly mean in the studied basin, based upon the performance of the model. The ANN architecture that produced the better result was RC315L with values for the determination coefficient, efficiency coefficient and standard estimate error (SEE) equal to 92.0, 77.0% and 8.29 respectively.