Comparison of AMS and POT Analysis with Long Historical Precipitation and Future Change Analysis Using “d4PDF”

Abstract

This paper assesses extreme precipitation events, which are one of the most impactful hydrological circulation events for policy decisions. In extreme analysis, samples of block maxima or peaks over threshold (POT) are used. However, annual maximum series (AMS), a type of block maxima, has been employed more often than POT. This study deals with problems that often occur in extreme analysis with long historical records and large ensembles of climate simulations. The following work is carried out: (1) AMS analysis is compared with POT analysis using long-term historical precipitation records at meteorological stations in Tokyo and Nagoya. With a carefully selected threshold, the generalized Pareto distribution keeps a more stable shape parameter than the generalized extreme value (GEV) distribution for AMS and gives relatively reliable return levels along with accumulated observation; and (2) a method using 60-year maxima is introduced to manage a very large set of AMS samples to which both the Gumbel distribution and the GEV distribution cannot fit well. Figures to obtain the 100-year return level are prepared based on the Gumbel distribution through the examination of relationships among past and future statistics of 1- to 3-day precipitation with a large ensemble of climate simulations called d4PDF. It is interesting that there are common statistical characteristics among the 1-day, 2-day, and 3-day precipitations.