Streamflow modelling is a very important process in the management and planning of water resources. However, com-plex processes associated with the hydro-meteorological variables, such as non-stationarity, non-linearity, and randomness, make the streamflow prediction chaotic. The study developed multi linear regression (MLR) and back propagation neural network (BPNN) models to predict the streamflow of Wadi Hounet sub-basin in north-western Algeria using monthly hy-drometric data recorded between July 1983 and May 2016. The climatological inputs data are rainfall (P) and reference evapotranspiration (ETo) on a monthly scale. The outcomes for both BPNN and MLR models were evaluated using three statistical measurements: Nash–Sutcliffe efficiency coefficient (NSE), the coefficient of correlation (R) and root mean square error (RMSE). Predictive results revealed that the BPNN model exhibited good performance and accuracy in the prediction of streamflow over the MLR model during both training and validation phases. The outcomes demonstrated that BPNN-4 is the best performing model with the values of 0.885, 0.941 and 0.05 for NSE, R and RMSE, respectively. The highest NSE and R values and the lowest RMSE for both training and validation are an indication of the best network. Therefore, the BPNN model provides better prediction of the Hounet streamflow due to its capability to deal with complex nonlinearity procedures.
en
pl
