Separation of Skewness: Reality of Regional Artifact?

Abstract

Statistical distributions have often been used to estimate future flood probabilities based on a past record of flood events. To choose such distributions, one study in the United States considered a number of hydrologically homogeneous regions and examined the coefficient of skevvness of annual flood series in these regions. It concluded that the skewness of flood series was, for each of the regions, more variable than that calculated from data sets generated randomly from a group of commonly used distributions. It referred to this phenomenon as the condition of separation (of skewness) or separation effect. Some authors have set a criterion that for a distribution to be adequate for flood frequency analysis it must be able to explain this separation phenomenon. We examine this problem and show that one primary cause of this phenomenon is the spatial mixing of skewness values within the same region. We conclude that one is not justified in using the separation of skewness as a criterion for choosing the type of distribution to be used in a flood frequency analysis.