Flood frequency analysis (FFA) entails the estimation of the uppertail of a probability density function (PDF) of annual peak flowsobtained from either the annual maximum series or partial durationseries. In hydrological practice, the properties of various methodsof upper quantiles estimation are identified with the case of knownpopulation distribution function. In reality, the assumedhypothetical model differs from the true one and one cannot assessthe magnitude of error caused by model misspecification in respectto any estimated statistics. The opinion about the accuracy of themethods of upper quantiles estimation formed from the case of knownpopulation distribution function is upheld. The above-mentionedissue is the subject of the paper. The accuracy of large quantileassessments obtained from the four estimation methods is compared totwo-parameter log-normal and log-Gumbel distributions and theirthree-parameter counterparts, i.e., three-parameter log-normal andGEV distributions. The cases of true and false hypothetical modelsare considered. The accuracy of flood quantile estimates depends onthe sample size, the distribution type (both true and hypothetical),and strongly depends on the estimation method. In particular, themaximum likelihood method loses its advantageous properties in caseof model misspecification.
展开▼
