Effects of estimation techniques on generalised extreme value distribution (GEVD) parameters and their spatio-temporal variations

The application of generalised extreme value distribution (GEVD) requires the estimation of three parameters. Different researchers adopted different techniques for the estimation of the GEVD parameters and no standard comparison amongst those methods are available. This paper investigates the comparison of the commonly used GEVD parameters’ estimations for extreme rainfall modelling. The maximum likelihood estimation, generalised maximum likelihood estimation, Bayesian and L-moments methods were considered in this study to compare the magnitude of the GEVD parameters and the corresponding return level estimations. The analysis was performed using the monthly and yearly extreme rainfall of Tasmania, Australia. The GEVD was fitted to four different data sets using the four parameters estimation techniques. Estimated return levels of the GEVD for all the estimation techniques were compared with the return levels provided by the Australian Rainfall and Runoff (ARR), which is the national guideline for Australian rainfall and flood studies. The outcomes of the analysis suggest that the L-moments method is the better estimator of the return levels when comparing the ARR provided return levels.