Abstract: Many methods are available to identify different components of a hydrologic time series, but differences in their performances are not understood clearly. Based on the probability theory, this paper describes two types of error evaluation standards for comparison of the consistency test accuracies assessed by five measures–Hurst measure, Bartels measure, Runs measure, Ljung-Box Q (LBQ) measure, and Spearman measure. Analysis results of the synthetic series, generated by the Monte-Carlo method, show the performances of all these measures are mainly affected by different variation types and different variation degrees, depending little on the mean, variation coefficient, or skewness coefficient of the stochastic components. Comparatively, Bartels measure, Runs measure and LBQ measure perform best, followed by Spearman measure, while Hurst measure performs worst. Application to five observed hydrologic time series indicate all the five measures have certain errors due to significant variability in the series but can achieve accurate assessments of the consistent parts of the series. Thus, we suggest that for higher accuracy, the two components of a time series, variation part and consistent part, be identified separately, with the variation part to be identified first and then the rest to be calculated for consistency test.

Key words: hydrological variation, consistency measure, Monte Carlo, correlation coefficient, statistical experiment