Research on flood frequency analysis based on partial duration series (PDS) has in recent years mainly focussed on the use of the generalized Pareto (GP) distribution for modeling exceedances. Davison and Smith [1990] mention several advantages of this distribution for use with PDS flood data. Moreover, they provide an interesting discussion of theoretical aspects of the PDS method along with recommendations as to its practical use, including the incorporation of additional information through covariates [see also Barret 1992]. The estimation of the GP parameters and corresponding sampling properties have been addressed among others by Hosking and Wallis [1987], Wang [1991], Rosbjerg et al. [1992], and Moharram et al. [1993]. Rosbjerg et al. [1992] concluded that for many typical applications in hydrology, the classical exponential model may yield better quantile estimates in terms of mean square error than the Pareto model, because there is one parameter less to be estimated (an analogous conclusion was reached by Lu and Stedinger [1992] ). Recent studies of the prediction precision of the exponential model comprise Rasmussen and Rosbjerg [1989, 1991], and Buishand [1989, 1990]. Ekanayake and Cruise [1993], on the other hand, found that the Weibull distribution was preferable to the exponential distribution.
Little has been done to include PDS models in a regional estimation scheme, and this is perhaps the main reason that PDS analysis remains less used in practice than the annual flood method. Future research should focus on developing regional estimation procedures for use with PDS data.