Evaluation of Parameter Estimation Methods for Flood Frequency Analysis: Computer Programs

摘要

Computer programs were used to assess the statistical performance indices of various estimators available for the flood frequency distribution under consideration. This involved three main phases of development: (1) generation of the pseudo-random numbers for the respective distribution; (2) estimation of parameters and quantiles by each of the available methods; and (3) computation of performance indices of bias, standard error, and root mean square error of the estimator. The generation of random numbers for the extreme value type I (EVI) and the two-component extreme value distributions is straightforward. Several alternative generation schemes were tested for generating the log Pearson type 3 (LP3) numbers. The schemes were found most suitable (SUBROUTINE LP3GN4) consisted of generating one parameter gamma variates using IMSL routine and transforming them to LP3 numbers. Each of the available estimation methods for a given distribution was programmed as independent subroutine(s). After obtaining the parameter and quantile estimates from each of the estimators, their performance indices were calculated in a straightforward manner. The user can easily combine the various subroutines to obtain the parameter and quantile estimates by various estimators in a single run.