A Unified Approach to Flood Frequency Analysis

Most stochastic hydrologic studies are based on statistical distributions that are pre-selected without giving consideration to the applicability of the distribution to a particular data series. In the United States for example, the Log-Pearson Type III distribution is suggested for national use (Water Resources Council, 1967). Many regulatory entities require that a particular distribution be used for statistical analyses, regardless of the goodness-of-fit. Statistical tests for goodness-of-fit (e.g. Chi-Square, Kolmogorov-Smirnov) can easily be employed to test the theoretical hypothesis that a particular distribution fits a given data series. It is logical and, now with the aid of computers, relatively simple to reduce a data series using several different statistical distributions, and to compare the applicability of all distributions based on a goodness-of-fit test. The distribution passing the goodness-of-fit test by the widest margin can be used as the 'best' fit of the data series.