Muti GCM/RCM ensemble dynamical downscaling experiments for Japan.

  • Suzuki-Parker, A., I. takayabu, H. Kusaka, K. Dairaku, S. Ham, S. Adachi, and N. Noriko, 2014: Muti GCM/RCM ensemble dynamical downscaling experiments for Japan. AOGS 2014, 118, Sapporo, Japan. 2014/08/01(謝辞:創生)→詳しくはこちら .

Dynamical downscaling (DDS) is an important tool for obtaining fine-scale climate information. There are a number of sources of uncertainty in DDS, including selection and configuration of regional climate models (RCMs), boundary forcing (namely reanalysis and/or course-resolution global circulation models), and their combined effects. Qualitative assessment of uncertainty contribution from each of these sources is crucial not only for reliable climate projection but also for effectively sampling the range of uncertainty. Assessments of uncertainty contribution from GCMs and RCMs can be and have been done with existing DDS ensemble experiments, such as EMSENBLES, NARCCAP, and CORDEX series. Previous studies based on these projects suffered from lacking ensemble members, as not all GCMs were downscaled with all RCMs (note that they are currently aiming to fill their GCM?RCM matrices). The authors are currently conducting multi GCM/RCM DDS simulations for Japan. The ensemble matrix is comprised of three GCMs (MIROC5, MRI-CGCM3, and CCSM4, under historical and RCP4.5 scenarios) and four RCMs (MRI-NHRCM, NIED-RAMS, Tsukuba-WRF, and AORI-RSM). RCM domain and resolution are common to all RCMs; Japan and its vicinity at 20km horizontal resolution. A distinctive feature of this ensemble experiment is the “filled matrix”, where all of the three GCMs are downscaled by all participating RCMs. The size of the matrix is small yet sufficient for applying Analysis of Variance (ANOVA) to quantitatively assess uncertainty attribution from GCMs and RCMs. Another key point is an objective selection of GCMs. We applied a cluster analysis to tropical SST change (which is known to be strongly influential to regional scale climatology) in CMIP5 GCMs, and selected one GCM from each cluster groups. The talk will focus on the simulated seasonal mean surface temperature and precipitation, as well as some discussions on the range of uncertainty arising from GCM and RCM differences.