Intercomparison of regionalization approaches for extreme rainfall modeling

We discuss the performances of two different approaches in evaluating extreme rainfall distributions from empirical rainfall records. The first approach fits a generalized extreme value (GEV) distribution to the empirical annual maxima. The second approach makes use of a larger portion of information and fits a Generalized Pareto (GP) distribution to the rainfall excesses above a range of suitable intensity thresholds u. We call this the “multiple threshold method” (MTM). The analysis is conducted using a database of 217 daily rainfall records from Sardinia (Italy) during the period 1922 to 1996, each having more than 40 years of observations. The performance of each approach is evaluated using a jacknife-type technique in order to estimate and compare the accuracy at ungauged sites. Initially, we fit the two distribution models independently to all available records, excluding one site per time. Then, we infer the distribution parameters at the excluded sites either through kriging or by assuming constant parameter values inside statistically homogeneous sub-regions. Finally, we discuss the results of the two approaches in terms of bias, variance and root mean square error, using the differences between the empirical annual maxima and those calculated theoretically from the fitted distribution models and the empirical exceedance probabilities of the ranked maxima.