A comparative evaluation of the estimators of the log Pearson type (LP) 3 distribution

Abstract

The log Pearson type 3 (LP3) distribution, recommended by the U.S. Water Resources Council (USWRC) in 1967, and subsequently updated in 1975, 1977 and 1981 as the base method of flood frequency analysis in the United States, has been widely used in many parts of the world. However, the estimation procedure for the LP3 distribution recommended by the USWRC has been shown by many investigators to perform rather poorly. In this study several estimation methods for the LP3 distribution, some quite recent, are investigated. The performance of the various methods is then compared via Monte Carlo simulation with the objective of identifying the most robust estimator among them. It is found that besides the USWRC method, the methods based on maximum likelihood and entropy perform poorly. The method based on the moments in real space, and the relatively recent so-called method of mixed moments perform markedly superior to other methods in terms of both resistance and efficiency of estimation or robustness. Therefore, the USWRC guidelines are in need of revision.