A COMPARISON OF TAIL PROBABILITY ESTIMATORS FOR FLOOD FREQUENCY-ANALYSIS

Selected techniques for estimating exceedance frequencies of annual ma ximum flood events at a gaged site are compared in this paper. Four ta il probability estimators proposed by Hill (PT1), Hosking and Wallis ( PT2) and by Breiman and Stone (ET and QT), and a variable kernel distr ibution function estimator (VK-C-AC) were compared for three situation s - Gaussian data, skewed data (three-parameter gamma) and Gaussian mi xture data. The performance of these estimators was compared with meth od of moment estimates of tail probabilities, using the Gaussian, Pear son Type III, and extreme value distributions. Since the results of th e tail probability estimators (PT1, PT2, ET, QT) varied according to t he situation, it is not easy to say which tail probability estimator i s the best. However, the performance of the variable kernel estimator was relatively consistent across the estimation situations considered in terms of bias and r.m.s.e.