Modeling low flows by univariate hybrid wavelet-time series approach

Abstract This research addressed the forecast of 7-day and 30-day low flows using daily flow data at two hydrometric stations of Qaleh-Shahrokh and Eskandari in Isfahan province, Iran. The two methods of ordinary time series modeling and time series modeling with the help of wavelet (wavelet-time series) are used to forecast low flows. The research also employs the wavelet method as a modern and effective way for the analysis of hydrological time series. In the wavelet-time series modeling, the target time series is decomposed into five levels using the Haar wavelet theory. With this decomposition, the series is divided into two parts including approximation a, which is the main nature of the data, and detail sub-series, which includes the white noise of the data. Then, the time series modeling steps are implemented for the approximation a. The results of the forecast of 7-day and 30-day low flow using the time series and wavelet-time series methods are explored by the error assessment criteria including the coefficient of correlation between the predicted and observed values, root mean squared error (RMSE), and mean absolute deviation (MAD). Finally, the coefficients of correlation between the predicted and observed values of 7-day and 30-day low flows in the Qaleh-Shahrokh and Eskandari stations are estimated at 0.87, 0.87, 0.55, and 0.95 by the time series method and 0.99, 0.99, 0.99, and 0.99 by the wavelet-time series method, respectively. The results show that the wavelet-time series method outperforms the time series method in predicting 7-day and 30-day low flows.