Regional Frequency Analysis of extreme rainfall events, Tuscany (Italy)

Abstract:

The assessment of extreme hydrological events at sites characterized by short time series or where no data record exists has been mainly obtained by regional models. Regional frequency analysis based on the index variable procedure is implemented here to describe the annual maximum of rainfall depth of short durations in Tuscany region. The probability distribution TCEV - Two Component Extreme Value is used in the frame of the procedure for the parameters estimation based on a three levels hierarchical approach. The methodology deal with the delineation of homogeneous regions, the identification of a robust regional frequency distribution and the assessment of the scale factor, i.e. the index rainfall. The data set includes the annual maximum of daily rainfall of 351 gauge stations with at least 30 years of records, in the period 1916 - 2012, and the extreme rainfalls of short duration, 1 hour and 3, 6, 12, 24 hours. Different subdivisions hypotheses have been verified. A four regions subdivision, coincident with four subregions, which takes into account the orography, the geomorphological and climatic peculiarities of the Tuscany region, has been adopted. Particularly, for testing the regional homogeneity, the cumulate frequency distributions of the observed skewness and variation coefficients of the recorded times series, are compared with the theoretical frequency distribution obtained through a Monte Carlo technique. The related L-skewness and L-variation coefficients are also examined. The application of the Student t -test and the Wilcoxon test for the mean, as well as the χ² was also performed. Further tests of subdivision hypotheses have been made through the application of discordancy D and heterogeneity H tests and the analysis of the observed and the theoretical TCEV model growth curves. For each region the daily rainfall growth curve has been estimated. The growth curves for the hourly duration have been estimated when the daily rainfall growth curve is not able to describe the sample cumulate frequency distribution. In order to define the index rainfall for each homogenous region and each duration, a multivariate model that combines the Mean Annual Precipitation, the elevation, the aspect and the sample mean of the annual maxima time series of the gauge sites has been used.