Low flow frequency analysis based on parametric and nonparametric statistics

A currently used approach to low flow frequency analysis is based on the concept of parametric statistical inference. In this analysis the assumption is made that the distribution function describing the annual minimum low flow data is known, which is never known exactly. In recent years nonparametric methods of estimating probability distribution functions have been developed, which do not require a distributional assumption. Each of these involves the use of a suitable smoothing function known as a kernel. The fixed and variable kernel nonparametric methods are proposed and developed for estimating low flow quantiles. Based on annual minimum low flow data and Monte Carlo simulation experiments, the proposed models are compared with Weibull models both for their descriptive and predictive ability. Computed results showed that the fixed kernel estimator has small bias and root mean square error in low flow quantile estimates. Application of the models to data from the Blue Nile, Ruvu, Kiwira and Komati rivers have shown that the nonparametric approaches are viable alternative to the Weibull models. It is concluded that nonparametric methods are accurate, uniform, and particularly suitable for the multimodal data