Evaluation of the efficiency of custom and computerized methods for reconstruction of monthly flow time series in the hydrometric stations

The lack of complete data should not be the cause for disregarding the hydrological condition and the long-term forecasts for performing a hydrological project in one region. Therefore, various researchers have used different methods such as Ratio Analysis, Fragment, and Thomas-fiering for the reconstruction of incomplete flow data in hydrometric stations. So, in this study, the accuracy of these methods and computerized methods such as, artificial neural network, hybrid wavelet-neural network and support vector machine have been investigated and compared. The results showed that the computerized methods have the higher accuracy than the other three methods. Comparison amongst the computerized methods shows that the artificial neural network method (R^2=0.98,RMSE=6.18,SE=0.476), the support vector machine method (R^2=0.902,RMSE=6.074,SE=0.486) and the hybrid wavelet-neural method (R^2=0.889,RMSE=6.96,SE=0.54) ranking first, second and third, respectively. Although, these three methods of artificial neural network, hybrid wavelet-neural network and support vector machine have not significant difference in comparison with each others, but the support vector machine constructed the data in the less time and with the more ease and hence has an advantage in comparison with the other methods.