Regional Frequency Analysis of Precipitation Using Time Series Clustering Approaches

Abstract

A regional frequency analysis using L-moments was performed with time series clustering approaches to identify homogeneous regions using dynamic data sets in Northern Cyprus. In this context, the conventional approach, based on station characteristics and different time series clustering approaches, classified as shape-based, feature-based, and model-based, were compared. Hierarchical Ward’s method with the correlation-based similarity measure of the feature-based approach was determined as the best method regarding the results of the jackknife validation procedure, which was performed for assessment of clustering approach uncertainty. Therefore, the cluster analysis ended up with five homogeneous subregions, and according to the goodness-of-fit measure, the Pearson Type III, generalized logistic, and generalized normal distributions were chosen as the best fit for different subregions. The accuracy of the estimated quantiles was evaluated through Monte Carlo simulations and, consequently, the quantiles for different return periods were estimated, which demonstrated spatial consistency in terms of increasing trend from the low-lying Mesaoria Plain toward the north coastal strip, including the Kyrenia Mountains and the Karpass Peninsula.