Rainfall frequency analysis using block maxima and peaks over threshold approaches.

Abstract

In the present study extreme rainfall frequency analysis was performed using block maxima (BM) and peaks over threshold (POT) approaches based on daily rainfall data gathered from three meteorological stations in Greece (Larisa, Agxialos, Trikala). In the first method, 9 different probability distributions (2 and 3 parameter distributions) were fitted to the samples (that were previously checked for randomness, trends and change points in the mean) in order to estimate rainfall depths for high return periods. According to diagnostic plots and Kolmogorov-Smirnov goodness of fit test, the fitting was acceptable for all the distributions that were considered in the three stations. However, the 3-parameter distributions and more specifically GEV for Larisa and Trikala and 3-parameter Lognormal for Agxialos, had a better fitting in the extreme values and could be considered more suitable for statistical modeling in these three stations. In the POT approach a suitable threshold was selected based on plots and statistical indexes. The excesses were modeled using the Generalized Pareto (GP) and a mixed model of Poisson and GP distributions. The estimated rainfall depths in this approach show similarities between the two models, as well as with those of the 3 parameter distributions used in the BM approach.