An Evaluation of Artificial Neural and Neuro-Fuzzy Intelligent Models in Rainfall-Runoff Simulation: A Case Study of Balikhluchay Watershed, Ardabil Province, Iran

Due to different reasons like non-normative usage and sequential droughts, water scarcity has highlighted the necessity of water resources management. Studying the processes affecting water resources is necessary for a normative management with sustainable development. The most important hydrologic process is the rainfall-runoff process that influences both ground and surface water resources. The study of this process is closely related to water resources management. In the past two decades, the intelligent models have found more importance because of their high ability in simulation of nonlinear and complex processes like rainfallrunoff process. In this study, the efficiency of common intelligent models in hydrology including Multi-Layer Perceptron artificial neural network (MLP), Radial Basis Function artificial neural network (RBF) and Co-Active Neural Fuzzy Inference System (CANFIS) in rainfall-runoff simulation and estimation of monthly discharge in Balikhluchay watershed was investigated. According to the coefficients of MSE, NMSE, MAE and RMSE (0.0145, 0.276, 0.103 and 0.120 respectively), the MLP artificial neural network which has three hidden layers, three neurons in each layer, a momentum algorithm and hyperbolic tangent transfer which would stimulate the monthly discharge for a 24 month test period with high accuracy was chosen as the most accurate network.