GOODNESS-OF-FIT AND STABILITY ASSESSMENT IN FLOOD FREQUENCY ANALYSIS

Abstract

This paper describes the results obtained from flood frequency analysis in river basins in Japan. Using annual maximum discharge data at ninety-nine locations on the major rivers, the authors fitted a number of frequency analysis models (probability distributions) and evaluated them in terms of not only goodness of fit and but also stability of quantile estimates. The standard least-square criterion (SLSC) is used as a goodness-of-fit criterion. The stability of quantile estimates is assessed in terms of estimation error obtained by the jackknife method, which can be used for bias correction and quantification of estimation error. Based on the application results for ninety-nine samples, this paper has revealed:(1) goodness of fit of various models;(2) that SLSC =0.04 could be a threshold for good-fitted models; and (3) that the jackknife estimation error can be an index to evaluate the stability of the models. Importance of graphical analysis using probability paper is stressed.