USE OF ARTIFICIAL NEURAL NETWORKS FOR SIMULATING ADHAIM RIVER BASIN, IRAQ

Adhaim River is one of Tigris river tributaries, contributing by 5555% of the Tigris River yield, with annual flow rate of 0.832 BCM. To simulate Adhaim River flow, Artificial Neural Networks (ANNs) was used, 21 years' hydrological and meteorological data were collected and analyzed from Iraqi ministry of water resource. Catchment delineation was estimated using Geographical Information Systems (GIS( technique, to obtain the best simulation of the runoff, five different methods for estimated average rainfall and five formulas for scaled all data was examined. ANNs technique parameters was obtained by the most appropriate performance criteria of graphical and statistical approaches, such as number of neurons, layers, and epoch values. ten types of transfer functions, different learning rate. As a result, four models were developed with different time steps (i.e., 15, 20, 25, and 30 days). The study shows that the most appropriate ANNs algorithm was Levenberg-Marquradt with back propagation using one hidden layer and three transfer functions namely 'tansig', 'logsig', and 'trainlm', The networks' performance varied with different time step involved in the study; however, the 30 days was almost better than other networks.