A cautionary note on decadal sea level pressure predictions from GCMs

• Published in: Advances in climate change research Vol. 9; no. 1; pp. 43 - 56
• Main Authors: Liess, Stefan, Snyder, Peter K., Kumar, Arjun, Kumar, Vipin
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.03.2018; Department of Earth Sciences, University of Minnesota, Minneapolis MN 55455, USA%Department of Soil, Water, and Climate, University of Minnesota, St.Paul MN 55108, USA%Computer Science & Engineering, University of Minnesota, Minneapolis MN 55455, USA; KeAi Communications Co., Ltd
• ISSN: 1674-9278; 1674-9278
• DOI: 10.1016/j.accre.2018.02.002

A comparison of sea level pressure (SLP) trends in a subset of seven Coupled Model Intercomparison Project (CMIP) phase 5 general circulation models (GCM), namely decadal simulations with CCSM4, CanCM4, MPI-ESM-LR, FGOALS-g2, MIROC4h, MIROC5, and MRI-CGCM3, to their CMIP3 counterparts reveals an unrealistically strong forecast skill in CMIP3 models for trend predictions for 2001–2011 when using the 1979–2000 period to train the forecast. Boreal-winter SLP trends over five high-, mid-, and low-latitude zones were calculated over the 1979–2000 initialization period for each ensemble member and then ranked based on their performance relative to HadSLP2 observations. The same method is used to rank the ensemble members during the 2001–2011 period. In CMIP3, 17 out of 38 ensemble members retain their rank in the 2001–2011 hindcast period and 3 retain the neighboring rank. However, numbers are much lower in more recent CMIP5 decadal predictions over the similar 2001–2010 period when using the 1981–2000 period as initialization with the same number of ensembles. Different periods were used for CMIP3 and CMIP5 because although the 1979–2000 initialization is widely used for CMIP3, CMIP5 decadal predictions are only available for 30-year periods. The conclusion to consider the forecast skill in CMIP3 predictions during 2001–2011 as unrealistic is corroborated by comparisons to earlier periods from the 1960s to the 1980s in both CMIP3 and CMIP5 simulations. Thus, although the 2001–2011 CMIP3 predictions show statistically significant forecast skill, this skill should be treated as a spurious result that is unlikely to be reproduced by newer more accurate GCMs.