A regional flood frequency model is developed for estimating recurrence intervals of extreme floods. The regionalization procedure developed in this paper differs notably from the U.S. Geological Survey index flood method (Dalrymple, 1960) in that a large quantile, rather than the mean annual flood, is used as the “index flood.” Based on a result from extreme value theory, exceedances of the specified quantile are modeled by a generalized Pareto distribution. The generalized Pareto distribution has two parameters: a scale parameter and a shape parameter. It is assumed that the shape parameter does not vary from basin to basin, implying that annual peak distributions for all basins have the same upper‐tail thickness. The scale parameter, on the other hand, not only varies from basin to basin, but may also depend on covariate information, such as drainage area or indicator functions for basin geology. Likelihood‐based inference procedures are developed for the regional flood frequency model. The model is applied to extreme floods of the Central Appalachian region of the United States.
All Science Journal Classification (ASJC) codes
- Water Science and Technology