Estimating probability of extreme rainfall over Japan using extended regional frequency analysis

Abstract

A method of frequency analysis, Extended Regional Frequency Analysis (ERFA), is proposed for reliable estimates of extreme daily rainfall probabilities for a long return period from relatively short daily rainfall records. The method uses combined data in a wide meteorologically homogeneous region (e.g., all Japan) to ensure a large number (order of 10,000) of data to minimize the effects of statistical sampling error in the frequency analysis. We applied the ERFA to daily rainfall data observed over Japan and to a high resolution atmospheric model simulation data over the meteorologically homogeneous land region of Japan. We found very good agreement between the empirical probability distribution and theoretical distribution estimated by ERFA, suggesting that the method is promising. However, we have noted some problems regarding ERFA: selection of the distribution, selection of the region, and model bias. These problems, along with possible solutions, are discussed.