Nonparametric Estimation of Low-Flow Frequencies

A new nonparametric method is proposed for low-flow frequency analysis. Low-flow series often contain years with zero values. The occurrence of zero and nonzero events are treated by first estimating the probability of zero flows using a plotting formula, and then fitting the nonzero events by nonparametric method. A Monte Carlo–simulation experiment is performed to compare the proposed nonparametric method with two parametric distributions, namely, log-Pearson Type III and Weibull. Results show that the nonparametric method gives more accurate results than currently used parametric methods. It is also concluded that the nonparametric method is uniform because it does not require a priori specification of any parametric distribution.