simulation of taleghan reservoir daily inflow using hammerstein-wiener models

hammerstein-wiener (hw) models are capable in describing nonlinear dynamic systems. these models are nonlinear and have been widely used in a wide range of sciences due to their simplicity and having a physically-based concept. in this research, for the first time in hydrology and water resources management, three different structures of these models using daily temperature and precipitation data as model inputs were applied  to simulate taleghan reservoir daily inflow using r2, rmse, srmse, mae, d and pep statistics and criteria. to do this, the reservoir data from 2006 to 2011 were utilized. the results obtained with (hw1) and without (hw2) data pre-processing were compared with the results achieved from two different structures of artificial neural networks (anns) including (i) feed-forward ann with two hidden layers (feedf2) and (ii) generalized regression neural network (grnn2). the results revealed that the hw models outperformed the ann models. in particular, the mean and standard deviation of the inflow time series were simulated very accurately. the srmse values of the hw1 model were 33% and 37% and while these values for the hw2 model were 28% and 43% over calibration and validation phases, respectively. meanwhile, the accuracy obtained over calibration and validation phases were 50% and 71% for feedf2 and 58% and 50% for grnn2, respectively.