A Smirnov test-based clustering algorithm with application to extreme precipitation data

Abstract

A clustering algorithm is developed tO form regions with similar extreme rainfall cumulative distribution function (CDF) characteristics. Stations are grouped on the basis of minimum geographic distance and acceptance of the null hypothesis of equal CDF between all station pairs within a cluster. During each iteration previously clustered stations can be regrouped on the basis of the results of a suite of Smirnov tests. This process continues until all possible cluster mergers have been disallowed and thus the final number of clusters is determined solely by the grouping process. The Smirnov test-based algorithm is applied to extreme rainfall data from West Virginia. The results are compared based on the L moments heterogeneity measure. With minor exceptions, the resulting subregions were deemed homogeneous by this measure. Thus it is possible that the Smirnov test-based clustering procedure can be used as a guide for the otherwise subjective formation of precipitation regions that is a prerequisite of L moments distribution fitting routines.