Single neural network and neuro-updating conceptual model for forecasting runoff. (2016)

• Record Type: Journal Article
• Authors: Boulmaiz, Tayeb; Ouerdachi, Lahbassi; Boutoutaou, Djamel; Boutaghane, Hamouda
• Is Part Of: International journal of hydrology science and technology. Volume 6:Number 4(2016)
• Page Start: 344
• Page End: 358
• Publication Date: 2016
• Subjects: Algeria -- artificial neural networks -- ANNs -- conceptual models -- runoff forecasting -- GR4J -- hydrology -- NARXN model -- Oued Rassoul -- output updating -- rainfall runoff -- water resources -- modelling -- water management -- flood protection; Hydrology -- Periodicals; Water resources development -- Periodicals; Hydrology; 553.705
• Languages: English
• ISSNs: 2042-7808

Abstract

Application of artificial neural network (ANN) becomes an alternative or complementary technique for forecasting runoff, which is an important operation for water management or flood protection. Two using approaches of ANN are applied in this study. The first is to use a nonlinear autoregressive with exogenous inputs network (NARXN) as a single model with rainfall data as inputs to forecast runoff, while the second is to use a feedforward ANN to update the outputs of a global conceptual model (GR4J). Trial and error procedure has been used for both ANNs in order to search combinations that give the best accuracy. The results show that the second approach performs better for forecasting daily flow (mean square error in test period = 0.19) and is more efficient in terms of water resources with an annual mean supply error of 0.56 Hm 3 compared to NARXN model (5.29 Hm 3 ).