Comparative Evaluation of the Estimators of Some Flood Frequency Models Using Monte Carlo Simulation

摘要

Performance of estimators of several commonly used flood frequency models is evaluated in terms of the statistical criteria of bias (BIAS), standard error (SE), and mean square error (MSE). The procedure is based on Monte Carlo simulation. The models considered are the Gumbel's extreme value type 1 (EV1), log Pearson type 3 (LP3), and the two component extreme value (TCEV) distributions. The estimators used are based on the moments (MOM), maximum likelihood estimation (MLE), probability weighted moments (PWM), least squares (LEAS), incomplete means (ICM), mixed moments (MIX), and entropy (ENT). The performance of the EV1 estimators is evaluated for random as well as serially correlated samples. MLE provides efficient quantile estimates even for small samples, closely followed by ENT. The PWM quantile estimates are unbiased for random samples and least biased for serially correlated samples.