We analyzed the annual maxima series of daily precipitation data for 51 meteorological observatories, each having more than 100 records. After 12 probability distributions were fitted to data, their goodness-of-fit was evaluated using the standard least-squares criterion (SLSC). The 12 probability distributions include the Lévy distribution, which has a fat tail of probability density function, and which has not been used for frequency analyses of annual maxima in the Japanese hydrological community. Some researchers have demonstrated that using more parameters of probability yields better goodness-of-fit test results. A probability distribution with fewer parameters is generally preferred. However, we sought to estimate the return period of the largest recorded daily precipitation as less than several hundred years. Therefore, we used distributions with more than three parameters, especially the Lévy distribution, which has a fat tail. We used many distributions including the Lévy distribution. Therefore, we obtained return periods of less than several hundred years for all observatories. We estimated the 120-year daily precipitation for these 51 observatories and drew a contour map.
View full abstract