Development of sampling downscaling: a case for wintertime precipitation in Hokkaido

Abstract

This study has developed sampling downscaling (SmDS), in which dynamical downscaling (DDS) is executed for a few of period selected from a long-term integration by general circulation model based on an observed statistical relationship between large-scale climate and regional-scale precipitation. SmDS expectedly produces climatology and frequency distribution of precipitation over a nested region with reducing computational cost, if a global-scale climate pattern mostly controls regional-scale weather statistics. Here SmDS was attempted for wintertime precipitation over Hokkaido, Japan, because a linkage between snowfall and sea-level pressure patterns has been known by Japanese synopticians and it can be detected by singular value decomposition (SVD) analysis on wintertime inter-annual variability during the period from 1980/1981 to 2009/2010 for precipitation over Hokkaido and moisture flux convergence around there. DDS for the full period over the same domain was also performed for comparison with SmDS. SmDS selected two winters from the top and two winters from the bottom of the projection onto the first SVD mode. It was found that, comparing with the full DDS, SmDS indeed provided unbiased statistics for average but exaggerated extreme statistics such as heavy rainfall frequency. It was also shown that the sampling in the SmDS method was much more effective than the random sampling.