The assessment of artificial neural network rainfall-runoff models under different input meteorological parameters Case study: Seybouse basin, Northeast Algeria

Over the past two decades, artificial neural networks (ANN) have exhibited a significant progress in predicting and modeling non-linear hydrological applications, such as the rainfall-runoff process which can provide useful contribution to water resources planning and management. This research aims to test the practicability of using ANNs with various input configurations to model the rainfall-runoff relationship in the Seybouse basin located in a semi-arid region in Algeria. Initially, the ANNs were developed for six sub-basins, and then for the complete watershed, considering four different input configurations. The 1st (ANN IP) considers only precipitation as an input variable for the daily flow simulation. The 2nd (ANN II) considers the 2nd variable in the model input with precipitation; it is one of the meteorological parameters (evapotranspiration, temperature, humidity, or wind speed). The third (ANN IIIP,T,HUM) considers a combination of temperature, humidity, and precipitation. The last (ANN VP,ET,T,HUM,Vw) consists in collating different meteorological parameters with precipitation as an input variable. ANN models are made for the whole basin with the same configurations as specified above. Better flow simulations were provided by (ANN IIP,T) and (ANN IIP,Vw) for the two stations of Medjez-Amar II and Bordj-Sabath, respectively. However, the (ANN VP,ET,T,HUM,Vw)’s application for the other stations and also for the entire basin reflects a strategy for the flow simulation and shows enhancement in the prediction accuracy over the other models studied. This has shown and confirmed that the more input variables, as more efficient the ANN model is.