Uncertainty assessment of regionalized flood frequency estimates using a multivariate Gev model

The regionalization methods of flood frequency analysis are generally based on the assumption of independence in space of the maximum annual flood peaks. However, the generation of floods is associated with precipitation events characterized by a mesoscale or synoptic extension, so flood peaks at neighboring sites are cross-correlated to some degree. Because the amount of information from m correlated series is less than m times that from a single series, the available information in a region can be seriously affected by cross-correlation. This can introduce further uncertainties in flood regionalization. The present paper deals with a multivariate probability model with generalized extreme value marginals. The model can account for the presence of intersite correlation in the flood frequency regionalization exercise. Parameter estimation via maximum likelihood method is presented. The variance of quantile estimators is also determined through analytical derivation. Because of mathematical complexity of such derivation, an approximate formulation is introduced for practical computations. It represents an extension of the formula by De Michele and Rosso (2001) for the variance of normalized quantile estimators under the assumption of independent annual flood series in a region. This result is combined with the variance of index flood estimators to obtain an uncertainty model for evaluating the variance of flood estimators by the index flood method. Examples of application to northwestern Italy are presented and discussed.