This study aims at improving the accuracy for estimating probable extreme values with long return periods, such as 200 or 500 years, and this analysis utilizes annual daily maximum rainfall data observed in more than 100 years at 83 observatories of the Japan Meteorological Agency. At 71 observatories, the parameter
k, known as the GEV shape parameter, had a value from -0.2 to 0.0, and it had a tendency to converge in a narrower range or on a particular value as the sample number increases. In this study, the shape parameter
k was fixed to a certain value, which is equivalent to reducing GEV parameters from 3 to 2, and this procedure had an effect to reduce the unstableness of estimating probable extremes. It suggests this procedure will increase the reliability of probable extreme value estimation, for such observatories as having shorter observation periods, by applying the
k value at -0.1 as the nation-wide average value or at the average value obtained within a certain smaller region.
View full abstract