Comparisons of Weibull- and exponential-based partial duration stochastic flood models

A stochastic flood model with a Weibull distribution for flood exceedances is examined and compared to the traditional exponential-based model in terms of predictive and descriptive properties. The model is based upon the traditional representation of the flood mechanism as a Poisson counting process. It is expected that the more flexible Weibull distribution will be capable of modeling flood series which exhibit a wider range of variability than the exponential model. The Weibull-based model is shown to possess predictive properties which are superior to the exponential model when samples exhibit coefficients of variation less than 1.5 and sample sizes are on the order of 2 events per year. These characteristics are shown to exist in many observed flood series in the Gulf Coast of the United States.