Regionalization of Flood Data Using Probability Distributions and Their Parameters

摘要

The U.S. Geological Survey (USGS) recently used the method of residuals to delineate seven flood regions for the State of Kentucky. As an alternative approach, the FASTCLUS clustering procedure of the Statistical Analysis System is used to delineate five to six cluster regions in conjunction with statistical properties of the annual maximum floodpeak series, like the coefficient of variation as estimated using method of L-moments, L-moment ratio-coefficient of variation, the parameters of Extreme/Value Type-1 and Generalized Extreme Value flood frequency distributions, and the specific mean annual flood. For both cluster and USGS flood regions, regionalized flood frequency growth curves are developed and their performance evaluated using Monte Carlo simulation techniques. Flood regions are then evaluated and compared using trends in the hydrological characteristics of important variables, performance of the regionalized flood frequency growth curves, discriminant analysis and regression equations relating flood quantities to watershed physical characteristics.