Abstract: The Johnson distribution has many transformation forms, and therefore it can achieve a good curve fitting capability through complicated transformation curves. This article discusses parameter estimation using the moment method, quantile method, ordinary least square method (OLS), weighted least square method (WLS), and diagonal weighted least square method (DWLS). These five methods are used for curve fitting and the fitted Johnson distributions are compared with the fitting of P-Ⅲ distribution. To calculate precipitation frequency, we select the data series of annual precipitation at four meteorological stations (Dingbian, Yichuan, Zhidan and Wuqi) in northern Shaanxi, and use the Johnson distribution and the five parameter estimation methods to estimate the parameters of these data series. The results show that the annual precipitation frequency curve based on the Johnson distribution model can effectively fit the series; and DWLS gives the best fitting, followed by the moment method, OLS, WLS, and quantile method. In addition, the fitting of Johnson distribution is slightly better than that of P-Ⅲ distribution.

Key words: Precipitation, frequency analysis, Johnson distribution, P-Ⅲ distribution, parameter estimation