Evaluation of CMIP5 decadal experiments in prediction of SST modes of variability Can decadal prediction anticipate events such as the warming hiatus? Danny Barandiaran, Shih-Yu Wang Utah State University
The “warming hiatus”
r 2 < 0?
No slowdown in increase of ocean heat content Nuccitelli et al. 2013, Abraham et al No change in TOA radiation budget Trenberth and Fasullo 2010, Trenberth et al Decadal variability mostly confined to troposphere and surface ocean Forcing attributed to AMO (Liu, 2012) and more prominently to PDO/IPO (Meehl et al., 2011 & 2013; Dai et al., 2015)
CMIP5 Decadal Experiments Initialized with observations and run free for years Experiments start 1960, initialized at least every 5 years through 2005 (some through 2010) Prior work has shown the CMIP5 models exhibit PDO and AMO patterns, and decadal experiments show modest forecast skill (Smith et al, 2015, Kim et al. 2012)
CMIP5 Decadal Experiments For each decadal run, analysis considers forecast lead time year by year 18 Models used in analysis: BCC-CSM1.1, CanCM4, CCSM4, CFSV2-2011, CMCC- CM, CNRM-CM5, EC-EARTH, FGOALS-g2, FGOALS-s2, GEOS-5, GFDL-CM21, HadCM3, IPSL-CM5A-LR, MIROC4H, MIROC5, MPI-ESM-LR, MPI-ESM-MR, MRI- CGCM members for each model
So far, so good…
Looking for SST variability in CMIP5 We will define PDO, AMO, ENSO using area-averages of surface air temperatures, rather than EOF analysis on SSTs. “Physical” standpoint: problems with EOFs Practical standpoint: SST data for decadal runs not widely available
BCC-CSM1.1 CanCM4 CCSM4 GFDL-CM21 MIROC5 MPI-ESM-LR Kaplan SST
Skill Ensemble CanCM4 CCSM4 FGOALS-s2 GEOS-5 IPSL-CM5A-LR
Skill Ensemble CanCM4 CCSM4 FGOALS-s2 GEOS-5 IPSL-CM5A-LR
CMCC-CMS CNRM-CM5 GFDL-CM21 HadCM3 MIROC5 MRI-CGCM3 Kaplan SST
Skill Ensemble FGOALS-s2 HadCM3 IPSL-CM5A-LR MIROC5 MIP-ESM-MR
Skill Ensemble FGOALS-s2 HadCM3 IPSL-CM5A-LR MIROC5 MIP-ESM-MR
BCC-CSM1.1 CanCM4 CMCC-CMS CNRM-CM5 HadCM3 MIROC5 Kaplan SST
Skill Ensemble CFSv GFDL-CM21 HadCM3 MIROC5 MRI-CGCM
Skill Ensemble CFSv GFDL-CM21 HadCM3 MIROC5 MRI-CGCM
Tglb: Observed vs. Modeled
Conclusions CMIP5 decadal runs show some potential in prediction of events such as hiatus. Attempts to refine model ensemble via skill in decadal- scale SST variation offer little improvement. Selecting PDO offers best improvement. Large biases in land- and ocean-only T2m.