International Workshop ADVANCES IN STATISTICAL HYDROLOGY May 2325, 2010 Taormina, Italy Botero and Francés., Comparison of high return period floodquantile estimated with upper bounded distribution functions 1 by Blanca Botero (1) and Félix Francés (2) (1) Facultad de Ingeniería y Arquitectura, Universidad Nacional de Colombia sede Manizales. Manizales, Colombia.(baboteroh@unal.edu.co) (2) Instituto de Ingeniería del Agua y Medio Ambiente, Universidad Politécnica de Valencia, Valencia, Spain (ffrances@hma.upv.es ) Three distribution functions with upper bound were used to estimate high return period floodquantiles of Mediterranean rivers. The parameters of the EV4 (Extreme Value with four parameters), LN4 (Lognormal with four parameters) and TDF (Transformed distribution function) probability distribution functions were estimated by Maximum Likelihood method, using the NonSystematic information available. The present paper compares the different floodquantile estimated with each one of the upper bounded distribution functions, and in addition with other two common distributions, the TCEV and GEV (Two Component extreme value and General extreme value). Results of the frequency analysis with NonSystematic information, using distribution functions with an upper bound were successful. Comparison among high return period floodquantile estimates shows a different performance between quantiles estimated with the different distribution functions. Keywords: Flood Frequency Analysis, Historical information, Palaeofloods, Upper Bounded distribution functions, PMF. Flood frequency analysis is one of the most common methods to estimate the design flood for hydraulic structures and for flood hazard/risk mitigation programs. Traditionally, extreme flood estimates have been associated with large dam projects or with the location of nuclear and other high vulnerable facilities. For some of these projects, the design criteria commonly include the Probable Maximum Flood (PMF) estimation. The PMF is the biggest flood physically possible at a specific catchment (Smith and Ward, 1998). It has a physical meaning and it provides an upper limit of the interval within which the decision maker must operate and design. Related to high return period quantiles estimation, flood frequency analysis has a well known drawback, as pointed out by Klemes (1993) and more recently by Merz and Blöschl (2008): the lack of available information about large events in a relatively short data series recorded systematically at a flow gauge station (from now, Systematic information). This fact involves the extrapolation of very high return period quantiles from data records which rarely exceed a hundred of years, producing quantile estimates with a high level of uncertainty. In the last decades, as a way to solve this problem, many authors as Jin and Stedinger (1989), Frances et al. (1994), Cohn et al. (1997), Martins and Stedinger (2001), England et al. (2003), Naulet et al. (2005), Reis and Stedinger (2005) and Merz and Blöschl (2008), have included historic and palaeoflood information (from now, NonSystematic information) in flood frequency analysis with very good results. Probability distribution functions with 2, 3 or 4 parameters have been used in extreme floods frequency analysis, with the common characteristic of having a positive skewness coefficient (γ x ) and no upper bound. The use of parametric distribution functions allows the increment of the return period of the requested quantile as much as it is required (obviously increasing at the same time its uncertainty). However, as the return period increases, with this kind of parametric distribution functions, the estimated quantiles increase too with no limit. Though, the question to pose at this point is: would it be possible to have a flood with