Annual runoff prediction of the upstream of Heihe River Basin, China

Abstract

A prediction method of annual runoff is proposed for the upstream Heihe River Basin. The Mann-Kendall test, order cluster analysis, variance analysis and R/S analysis methods were employed to investigate trends, mutations, periods and persistence of annual runoff time series, respectively. Based on various components, the linear superposition method was applied to forecast annual runoff. The Nash-Sutcliffe efficiency coefficient (NSE), the mean absolute percentage error (MAPE) and the percentage of pass (POP) were utilized to evaluate the simulation and prediction effects of annual runoff. Moreover, the calibration period was between 1958 and 2010, while the validation period was from 2011 to 2015. The results show that i) the annual runoff time series had a significant increasing trend (0.65×108 m3/10a), a probable mutation at 2006 and two significant periods (6 years and 22 years) in the calibration period; ii) the randomization of annual runoff time series was improved after removing the trend, mutation and periodic components, with the Hurst exponents 0.52; and iii) the simulation effects of calibration period (NSE=0.74, MAPE=6.6% and POP=90.6%) and the prediction effects of validation period (MAPE=6.5%, POP=80%) jointly indicate that the annual runoff prediction method is suitable in this river basin.