Modeling of daily rainfall sequence and extremes based on a semiparametric Pareto tail approach at multiple locations

Abstract

Summary A stochastic generation framework for simulation of daily rainfall at multiple sites is presented in this study. The limitations of a Gamma distribution-based Markov chain model for reproducing high-order moments are well-known, and the problems have increased the uncertainties when the models are used in establishing water resource plans. In this regard, this study attempted to develop a semiparametric model based on a piecewise Kernel-Pareto distribution for simulation of daily rainfall in order to further improve the existing model in terms of reproducing extremes, and in addition, the algorithm to reproduce the spatial correlation was combined. The proposed model can essentially be seen as a piecewise distribution approach constructed by parametrically modeling the tails of the distribution using a generalized Pareto and the interior by kernel density estimation methods. As a result, a Kernel-Pareto distribution-based Markov chain model has been shown to perform well at reproducing most statistics, such as mean, standard deviation, skewness and kurtosis. The proposed model provided a significantly improved estimate of design rainfalls for all the stations.