A.R. Jahangir; M. Raeini; M. Ziatabar Ahmadi
Abstract
Abstract
Flood is one of the destructive natural phenomena and being able to forecast it is of great importance. Simulation of rainfall-runoff and flood is difficult due to influence of several factors. So far, different methods have been suggested for their analysis. The aim of this study was to compare ...
Read More
Abstract
Flood is one of the destructive natural phenomena and being able to forecast it is of great importance. Simulation of rainfall-runoff and flood is difficult due to influence of several factors. So far, different methods have been suggested for their analysis. The aim of this study was to compare the efficiency of artificial neural networks (ANNs) in simulating rainfall-runoff process with HEC-HMS model. For this purpose, the Kardeh watershed which is located in northeast part of Great Khorasan province was chosen and based on several precipitation hyetographs and their runoff hydrographs (total of 450 data from 30 selected phenomena) the study was performed. Back-Propagation (BP) algorithm ANN was learnt to the data using sigmoid activation function. The criterion for selecting the network parameters in learning stage was producing the least RMSE in ANN outputs. Based on the SCS method and curve number (CN) the HMS model was performed. To evaluate the ANN performance, the simulated and observed data of total discharge and volume of runoff, peak discharges and peak times were compared. The results showed that based on Delta learning rule the multi layers Perceptron (MLP) network with 29 neurons, simulated the rainfall-runoff process with a high accuracy only in the middle (hidden) layer. The correlation coefficients of the total discharge and volume of runoff were found to be highly significant (r=0.98 and 0.99, respectively). The ANN model could significantly simulate the peak discharge and peak time values (r =0.98 and 0.83, respectively). By analyzing the HMS model performance, the correlation coefficients of the observed and simulated discharges and volumes of runoff were found to be 0.82 and 0.98, respectively. Also, the correlation coefficients of simulated peak discharges and peak times with this model were 0.97 and 0.70, respectively. By performing the T-test analysis at 99% confidence level no significant differences between observed and predicted data was observed. It can be concluded that although no significant differences was found between the two methods, however, the results of evaluated parameters showed that ANN predictions were more precise in comparison with those of HMS model.
Key words: Simulation, hydrologic model, Rainfall-runoff, artificial neural network, HEC-HMS model, Kardeh watershed.
