Selection of the best fit probability distribution in rainfall frequency analysis for Qatar

Abstract

Design rainfall is widely used in urban infrastructure planning and design such as culverts and urban drainage systems. In design rainfall estimation, one of the primary steps is the selection of a suitable probability distribution that fits the observed rainfall data adequately. This study examines the selection of the best fit probability distribution in design rainfall estimation. The annual maximum (AM) rainfall data from 29 rainfall stations in Qatar are used in this study. The rainfall record lengths of these stations are in the range of 24–49 years (average of 36 years). Fourteen different distributions and three goodness-of-fit tests (Kolmogorov–Smirnov, Anderson–Darling and Chi-squared) are considered. Based on a relative scoring method, the GEV distribution is found to be the best fit distribution. Results from bootstrapping and simulation analyses show that sample estimates of skewness of the AM rainfall series are subject to a higher degree of sensitivity to data length compared with standard deviation and mean as expected. Since the rainfall quantile estimates of higher return periods are greatly influenced by skewness, a longer data length is needed in reducing the uncertainty in rainfall quantile estimates for higher return periods, which is currently unavailable in Qatar.