水文頻度解析モデルの母数推定法と確率水文量の変動性-3母数対数正規分布について-

Abstract

There are many methods which estimate three parameters of the log-normal dis-tribution. The methods of maximum likelihood, least squares, moments, as well asthe Ishihara-Takase, Iwai improved and sextile methods are tested by using Monte-Carlo simulation. The authors evaluate these methods in terms of the mean andstandard deviation of bias in parameter estimates and quantiles (T-year events) ob-tained by each method. They also discuss the relationship between the sample sizeand the estimation error of quantiles. When the sample size is greater than about50, the method of maximum likelihood is considered the best. For a small sample, the methods conventionally used in Japan, such as Ishihara-Takase and Iwai im-proved methods, give better estimates than the other methods.