Comparative Study of Conventional and Computerized Reconstruction Techniques for Flow Time Series Data of Hydrometric Station

Abstract

One of the undeniable requirements in hydrological forecasting and water resources studies is the availability of reliable information. Due to the various reasons, time series data are not usually complete in those surveys, therefore; reconstruction techniques are highly required to complete the missing data. This research was undertaken to evaluate the efficiency of the computer-based methods namely artificial neural network, support vector machine, ARIMA, and ARMAX along with conventional reconstruction strategies of ratio analysis, Fragment, and Thomas-Fiering. As a case study, the monthly flow data of seven hydrometric stations in the Urmia Lake Basin were employed during a 15-year period. The results were then compared using the evaluation criteria of the correlation coefficient (R2), root mean square error (RMSE), standard deviation ratio (SDR), Nash-Sutcliffe efficiency (NSE), and standard error (SE). Based on key results, computerized methods had higher accuracy than conventional ones in data reconstruction. In terms of efficiency, among the computer-based methods, the support vector machine, ARMAX, artificial neural network, and ARIMA model were ranked from the first to fourth in missing data regeneration.